TOWARDS AGRICULTURAL TRANSFORMATION: FACTORS INFLUENCING

i

TOWARDS AGRICULTURAL TRANSFORMATION:

FACTORS INFLUENCING THE CULTIVATION OF HIGH

VALUE AGRICULTURAL PRODUCTS IN UGANDA

By

Ryan Currier Jayne

Thesis presented in partial fulfillment of the requirements for the degree of

Master of Science (Agricultural Economics) in the

Faculty of AgriSciences

at Stellenbosch University

Supervisor: Mrs. Lulama Ndibongo-Traub

March 2018

ii

Declaration

By submitting this thesis, I declare that the entirety of the work contained therein is my own

original work, that I am the sole author thereof (save to the extent explicitly otherwise stated),

that reproduction and publication thereof by Stellenbosch University will not infringe any

third party rights and that I have not previously in its entirety or in part submitted it for

obtaining any qualification.

Date: December 2018

Copyright © 2018 Stellenbosch UniversityAll rights reserved

Stellenbosch University https://scholar.sun.ac.za

iii

ABSTRACT

Growing global markets have created opportunities that much of sub-Saharan Africa has been

leveraging through the expansion of export diversification. The share of high-value

agriculture (HVA) in total exports out of sub-Saharan Africa has increased from 8.4% in

2001 to 10.2% in 2016, although it is believed this is far beneath SSA’s true potential. The

emergence of domestic and international markets for high value agricultural products

presents a real opportunity for growth and development, specifically for smallholder famers

by providing increased economic returns and marketing opportunities. It is becoming widely

recognized that various development indicators can improve if smallholder farmers are better

integrated into these markets. Using Ugandan household panel data, this study seeks to

understand the factors related to the decision to cultivate HVA and the households’ marketing

outcomes. A triple-hurdle model is employed to robustly examine market-related decisions

made by smallholder farmers beyond common approaches to market participation models.

Results indicate that policies that encourage HVA market participation simultaneously

increase the likelihood of non-producers of HVA to commence producing and lead to greater

levels of net sales in the market. Furthermore, HVA producers have greater likelihoods

associated with being net sellers in the market.


iv

Uittreksel

Groeiende internasionale markte het geleenthede vir ŉ beduidende gedeelte van Afrika suid

van die Sahara (SSA) geskep, hierdie is deur die groei in uitvoer diversifikasie versterk. Die

aandeel van hoë-waarde landbou (HWL) uitvoere in totale uitvoere in SSA het tussen 2001 en

2016 van 8.4% tot 10.2% gestyg, maar sommiges is steeds van mening dat hierdie onder die

streek se potensiaal is. Die totstandkoming van plaaslike en internasionale markte vir HWL

produkte bied ware geleenthede vir groei en ontwikkeling, spesifiek vir kleinboere deur die

skep van ekonomiese- en bemarkingsgeleenthede. Dit is nou alombekend dat verskeie

ontwikkelingsaanduiders verbeter kan word indien kleinboere beter met markte geïntegreer

word. Hierdie studie gebruik ŉ Ugandese huishoudelike-paneeldatastel om huishoudings se

besluit om HWL produkte te produseer beter te verstaan en die bemarkingsuitkomste daarvan

te kwantifiseer. In teenstelling met die meer algemene markdeelname-benadering gebruik

hierdie studie ŉ driedubbele-versperringsmodel om die markverwante besluite van

kleinskaalse boere op meer robuuste wyse te ondersoek. Die resultate dui aan dat die beleid

wat HWL mark deelname aanmoedig beide die waarskynlikheid van HWL produksie onder

nie-HWL kleinboere verhoog en tot hoër netto mark verkope lei. Verder het HWL produsente

ook ŉ groter waarskynlikheid om met netto verkopers in die mark geassosieer te word.


v

Dedication

This thesis is dedicated to my parents, Connie Currier, Thomas Jayne and Kimm Jayne


vi

ACKNOWLEDGEMENTS

I would like to express my gratitude to the following people for the support, encouragement,

love and expertise that I received during the completion of my thesis.

To:

My mentor, advisor and boss, Lulama Ndibongo-Traub for her unconditional support and

guidance for the duration of this project.

My mother Constance Currier for her endless care, support and love.

My father Thomas Jayne for his support and encouragement.

And Jillian Kuluse for her wonderful company and love.

As well as a sincere thank you to anyone who added value to my study experience through

comradery, support, challenge, and new perspective at Stellenbosch University.


vii

TABLE OF CONTENTS

Declaration................................................................................................................................ ii

Abstract ................................................................................................................................... iii

Uittreksel .................................................................................................................................. iv

Dedication ................................................................................................................................. v

CHAPTER 1: INTRODUCTION ........................................................................................... 1

1.1 Structural Transformation ................................................................................................ 1

1.2 Regional Patterns of Land Use and Crop Diversification ................................................ 4

1.3 Purpose of the Study and Research Questions ................................................................. 7

CHAPTER 2: THE UGANDAN CONTEXT ...................................................................... 10

2.1 Geographical Context ..................................................................................................... 10

2.2 Historical and Economic Context .................................................................................. 11

2.3 Smallholder Farming ...................................................................................................... 14

2.3.1 Diet and Consumption ................................................................................................. 15

2.3.2 Agricultural Diversity ................................................................................................. 16

2.3.3 Farming Systems ......................................................................................................... 19

CHAPTER 3: THEORETICAL FRAMEWORK .............................................................. 22

3.1 Crop Diversification ....................................................................................................... 22

3.1.1. Crop Specialization .................................................................................................... 22

3.1.2. Crop Diversification and Determinants...................................................................... 24

3.2 Agricultural household models ...................................................................................... 28

3.2.1 Basic Model................................................................................................................. 30

3.2.2 Multiple Crop Model ................................................................................................... 31

3.3 Previous Application of Crop Diversification Models ................................................... 33

3.3.1 Research Application .................................................................................................. 34

3.3.2 Review of Double-Hurdle Applications and Triple Hurdle Background ................... 35

CHAPTER 4: EMPIRICAL FRAMEWORK .................................................................... 38

4.1 Methodology: Triple-Hurdle Model............................................................................... 38

4.2 Econometric Estimation ................................................................................................. 46

CHAPTER 5: DATA AND VARIABLES DESCRIPTION .............................................. 53

5.1 Dependent Variables ...................................................................................................... 53

5.2. Explanatory variables .................................................................................................... 54

5.3. Descriptive Statistics ..................................................................................................... 59

5.4 Data ................................................................................................................................ 60

CHAPTER 6: RESULTS ...................................................................................................... 63


viii

6.1 Bivariate Relationship Analysis ..................................................................................... 63

6.2 Triple Hurdle-Model ...................................................................................................... 65

6.3 Policy Recommendations ............................................................................................... 76

CHAPTER 7: CONCLUSIONS ........................................................................................... 78

BIBILOGRAPHY .................................................................................................................. 81


ix

List of Tables

Table 2.1: Economic Overview for Uganda: 2017 .............................................................. 12

Table 2.2: Share of Agricultural Land Owned by Farm-holding Size.............................. 15

Table 2.3: Staple Food Consumption in Uganda ................................................................ 16

Table 2.4: Household Mean Share of Agricultural Land Disaggregated by Crop .......... 19

Table 2.5: Ugandan Crop Portfolios: 2010/2011 ................................................................. 20

Table 4.1: Expected Sign of Coefficients on Dependent Variable ..................................... 52

Table 5.1: Dependent Variables............................................................................................ 53

Table 5.2: Dependent Variables, Household and Regional Characteristics ..................... 59

Table 6.1: Correlations Between Dependent and Explanatory Variable ......................... 63

Table 6.2: Marketing Channels of HVA .............................................................................. 65

Table 6.3:Triple Hurdle Estimation ..................................................................................... 66

Table 6.4: Triple Hurdle Estimation (Parsimonious Model - Imputed Means) ............... 68

Table 6.5: Estimated Partial Effect on the Unconditional Expected Value of Net Sales. 73

Table 6.6:Estimated Average Partial Effect on the Probability of being a Net Selling

Producer of HVA ................................................................................................................... 74


x

List of Figures

Figure 1.1: Total Area Harvested in SSA: Disaggregated by Crop Category ................... 5

Figure 1.2: Effective Number of Crop Species for Select Countries in SSA ...................... 6

Figure 1.3: Production Orientation Options for Smallholder Farmers.............................. 7

Figure 2.1: Rainfall (mm) and Population Density (Km Squared) ................................... 11

Figure 2.2: Uganda Production, Yields and Population Trends ....................................... 14

Figure 2.3: Shannon Diversity Index - Uganda ................................................................... 17

Figure 2.4 Shannon Diversity Index - Selected Countries in SSA ..................................... 18

Figure 3.1: Triple-Hurdle Modeling Framework ............................................................... 37

Figure 4.1: Decision Pathways: Triple-hurdle Modeling Framework .............................. 39


xi

List of Abbreviations

Average Partial Effect (APE)

High Value Agriculture (HVA)

Inverse Mills Ratio (IMR)

Kilometer (Km)

Maximum Likelihood estimation (MLE)

Meters Above Sea Level (MASL)

Millimeter (mm)

Sub-Saharan Africa (SSA)

Ugandan Shillings (UGX)


1

CHAPTER 1: INTRODUCTION

Agriculture is recognized as the key sector to drive broad-based economic development

across sub-Saharan Africa. As such, the Comprehensive African Agriculture Development

Plan (CAADP) is a re-commitment by African Leaders to accelerated agricultural growth and

transformation. Under the Malabo Declaration in 2014, African policy-makers set the target

of halving poverty by 2025 through targeted commodity-specific policy interventions that

would drive inclusive transformation (AUC, 2014). Evidence of structural change for the

economy as a whole is proven to be catalyzed from within the agricultural sector for many

lesser developed countries (Shilpi and Emran 2016). Thus, to meet the objectives put forth

by African leaders in CAADP, it is imperative that we advance our policy maker’s

understanding of the factors associated with the decision of which crops to cultivate.

Smallholder farmers in much of sub-Saharan Africa cultivate a narrow range of crops with

low economic returns. The emergence of domestic and international markets for high value

agricultural products presents a real opportunity for growth and development, specifically for

smallholder famers by providing increased economic returns and marketing opportunities.

The share of high-value agriculture (HVA) in total exports out of sub-Saharan Africa has

increased from 8.4% in 2001 to 10.2% in 2016, although it is believed this is far beneath

SSA’s true potential. Policies supporting crop diversification could be a catalyst for – and are

at least necessary for agro-economic structural transformation that is known to accompany

long-run economic growth and poverty reduction.

Using Ugandan household panel data, this study seeks to understand the factors related to the

decision to cultivate HVA and the households’ marketing outcomes using a triple hurdle

model developed by Burke et al. (2015). Results indicate that policies that encourage HVA

market participation simultaneously increase the likelihood of non-producers of HVA to

commence producing and lead to greater levels of net sales in the market. Furthermore, HVA

producers have greater likelihoods associated with being net sellers in the market.

1.1 Structural Transformation

Historically, agriculture has played a changing role in the broader theories of economic

structural transformation. Agricultural transformation can be described as “the process by

which an agrifood system transforms over time from being subsistence-oriented and farm-


2

centered into one that is more commercialized, productive, and off-farm centered.” (AGRA,

2016). The classic models of structural change and economic growth with respect to Lewis

(1954), Kuznets (1973) and Timmer (1988) observe a declining share of agriculture in a

nation’s employment and GDP as a key feature of economic development and poverty

reduction (Alvarez-Cuadrado, 2011; Shilpi and Emran, 2016). Schultz (1953) states that

improved agricultural productivity is a necessary precondition for an industrial revolution

since increases in agricultural productivity raise per capita incomes, in turn generating

demand for non-farm activities (Ranis and Stewart, 1973; Mellor, 1976; Haggblade, Hazell

and Reardon, 2006; Alvarez-Cuadrado, 2011; Shilpi and Emrann, 2016). This sectorial

reallocation of resources is driven by Engel’s law, where greater per capita incomes from

increased agricultural productivity transfer labor from agriculture into other sectors of the

economy (Alvarez-Cuadrado and Poschke, 2011). The phenomenon of factor shifts out of

agriculture has played an important role in stimulating the process of agricultural and

economic transformation. Various researchers have sought to decompose these drivers on a

finer scale.

Five “interlinked” steps of agrifood transformations in the context of agricultural

transformation have been identified in Asia and are emerging in Africa. In Timmer’s (1988)

The Agricultural Transformation, he states that the evolving stages of an economy take a

remarkably uniform process across different countries, which manifest within the agricultural

sector. The steps include (1) urbanization; (2) diet change (3) agrifood system

transformation; (4) rural factor market transformation; (5) intensification of farm technology

(Johnston and Mellor, 1961; Ranis et al., 1990; Delgado et al., 1994; Timmer, 2002). These

represent, in macro terms, the fundamental drivers that occur as an economy becomes more

modernized.

The occurrence of rural to urban migration and urbanization is accompanied by lifestyle

changes. Through Bennet’s Law (Bennet 1954), when incomes rise, so does the desire for a

diversified diet, which can lead to changes in the product composition of demand (Reardon

and Timmer 2014). Typically, this can have several outcomes, which include growing

demand for meats, dairy products, fruits and vegetables, feed grains and vegetable oils,

processed foods for home cooking; and prepared foods consumed away from home, while

experiencing a shift away from cereals. Diversified diets may be a signal of crop

diversification if those crops are sourced domestically. However, Dawe (2015) points out


3

that dietary diversity at the national level need not imply crop or production diversity,

because international trade may be meeting domestic consumer demand for diverse products.

Crop diversification at the national-level is often determined to a large extent by the country’s

agro-ecological zones suitable for a broad range of crops, as well as its market development

and stage of agricultural transformation (Kurosaki 2003; Dawe 2015; Rao et al. 2006).

Contrarily, increased domestic demand for diverse products will serve as a catalyst for

product diversification if the region is suitable for those crops being demanded and market

conditions are conducive for such enterprises.

As the demand for a diverse range of goods incentivizes the sourcing of domestic products,

local supply chains transform, this development has taken the overarching term, “supply-

chain revolution” (Dawe 2015). Its process has been observed in two ways, first, as a

“Modern Revolution” via developments of large scale retail and second stage processing

growth. Secondly, as a “Quiet Revolution” which is catalyzed by small and medium-scale

farmers at the first-stage of processing and the provisioning of upstream agricultural services

(Reardon and Timmer 2007; Reardon et al. 2012a). To a large extent, these developments

have been fueled by increased incomes, urbanization, the participation of women in labor

markets, the onset of new processing technologies, and food diversification encouraged by

the retail sector (Gehlhar and Regmi 2005).

Subsequently, the availability and accessibility of factor markets to smallholder farmers make

the production of a diverse range of products possible. The rise of rural factor markets and

non-farm employment is a response to market growth and changing diets via urbanization

(Reardon and Timmer 2014). Rural factor markets include various on and off-farm actors in

the supply chain such as processors, wholesalers, transportation services, credit markets,

chemicals and machinery, among others. Their presence is integral in facilitating the

upstream linkages in the supply-chain which contributes to greater levels of off-farm

employment.

If the leaders of sub-Saharan Africa are committed to achieving transformation, then one of

the elements that warrants significant attention is the decision of which crops to cultivate.

Consistent with seminal definitions of crop diversification, a shift of resources from low

value agriculture to high value agriculture (Hayami and Otsuka, 1992; Vyas, 1996) must

occur. The progression of crop diversification is argued to not be limited only to production


4

processes, but changing marketing and commercially based activities that expand income

sources of smallholder farmers and grow the overall rural economy. Thus, crop

diversification can be an indicator of agricultural transformation as smallholder farmers

become more commercially oriented, as they shift away from staple, low-value agriculture to

higher valued crops.

The extensive body of literature on economic growth and agricultural transformation

(Reardon and Timmer 1988; Minot 2006; Gollin 2002; Kurosaki 2003; Ray 2010; Alvarez-

Cuadrado 2011) 1 contends that diversified enterprises play an important role in these

processes of agricultural transformation. Changes in land productivity are structurally related

to the reallocation of land use among different crops (Kurosaki 2003). Therefore, to improve

agricultural productivity, it is paramount to advance policy-maker’s understanding of the

factors that influence a farmer’s decision of which crops to cultivate.

1.2 Regional Patterns of Land Use and Crop Diversification

To date, staple crops, predominantly cereals, dominate small-scale farmers’ area harvested in

Africa. Figure 1.1 illustrates the total area harvested in millions of hectares for cereals, roots

and tubers, fruits (excluding melons) and vegetables in sub-Saharan Africa (SSA), from

1961-2014.

1 Timmer, Peter C., “The Agricultural Transformation,” in “Handbook of Development Economics,” Vol. 1, Amsterdam and New York: North Holland, 1988, chapter 8, pp. 275–331. Ray, Debraj, “Uneven Growth: A Framework for Research in Development Economics,”Journal of Economic Perspectives, 2010, 24, 45–60. Gollin, Douglas, Stephen L. Parente, and Richard Rogerson, “The Role of Agriculture in Development,” American Economic Review, Papers and Proceedings, 2002, 92, 160–164. Alvarez-Cuadrado, Francisco, and Markus Poschke. "Structural change out of agriculture: Labor push versus labor pull." American Economic Journal: Macroeconomics 3.3 (2011): 127-158. Kurosaki, Takashi. "Specialization and diversification in agricultural transformation: the case of West Punjab, 1903–92." American Journal of Agricultural Economics 85.2 (2003): 372-386.


5

Figure 1.1: Total Area Harvested in SSA: Disaggregated by Crop Category

Source: FAOSTAT 2017. Author’s own calculations

The share of cereals in total area harvested over the time period declined from 33.2% to

31.7%; tubers increased 6.1% to 7.0%; fruits increased from 3.2% to 3.6%; vegetables 0.9%

to 1.6%. In staple cereal production, yields can be variable and capital intensive while gross

profits are low (Davis 2006). In 2012, the Kenyan agricultural minister described the

situation of small-scale farmers saying, "Our farmers need to diversify their activities and

venture into horticultural farming instead of relying on maize and wheat where earnings are

low” (Kosgei 2012). This point reinforces that more smallholder farmers will grow their

earnings if they shift away from staples to higher valued crops.

Despite low levels of crop diversity at the continental level, crop diversity has been trending

upward in many sub-Saharan Africa countries, but at a slow rate. Figure 1.2 below illustrates

the state of play of eight countries in the region with regards to crop diversification. The

Shannon Diversity Index, expressed as the Effective Number of Crops Species (ENCS), is an

index for diversification that collectively includes eight sub-Saharan African countries from

1961-2013. The value of the index represents an estimate for the number of crop species

dominating production in the specific areas within sub-Saharan Africa. An increase of 42.3%

is realized over the past 50 years, showing a shift in the score from 7.55 to 10.75 which

indicates a substantial rise in the number of dominant crops under cultivation. In the early

0.00

20.00

40.00

60.00

80.00

100.00

120.00

140.00

160.00

19

61

19

63

19

65

19

67

19

69

19

71

19

73

19

75

19

77

19

79

19

81

19

83

19

85

19

87

19

89

19

91

19

93

19

95

19

97

19

99

20

01

20

03

20

05

20

07

20

09

20

11

20

13

Are

a H

arv

este

d (

In m

illi

on

s o

f h

ecta

res)

Year

Cereals Fruits exc melons Roots Tubers Vegetables


6

1990’s there is a perceivable increase in the ENCS score, as compared to previous decades.

Notwithstanding this upward trend, there remains great opportunity to increase crop diversity.

Figure 1. 1: Effective Number of Crop Species for Select Countries in SSA

(Countries include the Democratic Republic of the Congo (DRC), Malawi, Mozambique, South African

Tanzania, Uganda, Zambia and Zimbabwe) Source: Bureau for Food and Agricultural Policy (2016)

From a farmer’s perspective, the decision to diversify their crop mix is theoretically based on

a series of opportunities and threats, while controlling for other factors. Opportunities largely

stem from changing consumer demand, urbanization, export potential, and marketing

opportunities, to name a few (Krishi 2011). Strategies that address threats may include a

diversified crop portfolio to hedge against various risks, such as crop price and yield

fluctuations due to the effects of climate change or pests, among other variables.

Additionally, smallholder farmers might diversify to meet home consumption demand, then

sell their surplus in the market for added income. Such decisions can be wide-ranging and a

host of factors can play a role. Below, figure 1.3 provides an abstraction from Turner (2014),

which illustrates the different orientation that small-scale farmers may choose based on these

characteristics and their level of development.

0

2

4

6

8

10

12

19

61

19

63

19

65

19

67

19

69

19

71

19

73

19

75

19

77

19

79

19

81

19

83

19

85

19

87

19

89

19

91

19

93

19

95

19

97

19

99

20

01

20

03

20

05

20

07

20

09

20

11

20

13

EN

CS

Year


7

Figure 1.2: Production Orientation Options for Smallholder Farmers

Source: Turner 2014

1.3 Purpose of the Study and Research Questions

Limited research has made the link empirically between the decision to cultivate high value

agriculture (HVA) and marketing outcomes. Burke et al. (2015) implements a triple-hurdle

model that is the first of its kind to appropriately address such a research question for dairy

production. This method is used to examine whether policies encouraging market

participation may also induce non-producers of a good to commence production.

Furthermore, it allows the researcher to determine what factors influence the probability of an

agricultural household being a net-seller in the market, conditional on being a producer of a

given product of bundle of products. The value of using this methodology permits the

researcher to draw broader conclusions around the impact of policies that are aimed at

promoting market participation of smallholder farmers. It allows us to determine whether

policies aimed at facilitating market participation may have further reaching impacts on

smallholder welfare than prior research may have suspected. This is because policies

believed to promote market participation may also induce smallholders to start cultivating

HVA. A triple hurdle model is especially appropriate for goods that are less frequently

produced by the general population, such as HVA or dairy, for example.


8

Research Questions

If increased agricultural productivity, be it yields or growing rural incomes, is considered a

key indicator of agricultural transformation, this study argues that empirically, the cultivation

of higher valued agricultural products by smallholder farmers is an important factor in this

process. The importance of HVA is relevant since farmers who cultivate HVA are believed

to have greater likelihoods of becoming more commercially oriented, likely to sell greater

quantities in the market. Recently, the government of Uganda has implemented an array of

initiatives to promote smallholder commercialization and agricultural diversification. In the

past two decades, access to information on technology adoption and crop-specific training

from extension services should have particular relevance in encouraging a shift away from

staples to higher valued agriculture, which are expected to be significant determinants in this

study (Barungi et al. 2016; Komarek 2010).

The research objectives and hypotheses borrow Bill Burke et al. (2015) triple hurdle model

and use Uganda panel data from the Living Standards Measurement Study (LSMS) to

examine:

• What are the determinants of smallholder production of HVA

• What are the factors that determine whether a HVA producing smallholder is a net

buyer, autarkic, or net seller in the market

• How do these determinants affect the degree of market participation, or quantities

bought and sold, among participants?

This research hypothesizes that:

• Access to information, infrastructure, inputs, capital, and crop prices, are significant

determinants of HVA cultivation and market participation,

• The same processes that influence a farmer to cultivate HVA increase the probability

of being a net seller in the market

The triple-hurdle model allows for a more comprehensive analysis of how selected factors

influence farming and marketing decisions for a particular product beyond methods that have

been used before. The analyses control for household and district-level characteristics that

are widely cited factors which influence agricultural household decisions. Conventional

wisdom posits that these factors include but are not limited to household demographics,

capital and assets ownership, agro-ecological, crop prices and district-level features.


9

Chapter Two provides a glimpse into the historical context of Uganda along with a

smallholder overview, then proceeds with a literature review of crop diversification, the

empirical framework for the estimations, its results and conclusions.


10

CHAPTER 2: THE UGANDAN CONTEXT

2.1 Geographical Context

Located in the Rift Valley of Eastern Africa, Uganda sits in an area of semi-arid savannah,

bush and mountains (Quam 1997). Uganda is landlocked and densely populated with 206.9

people per square kilometer (World Data Atlas 2016). Today its total population is

approximately 41.49 million and is comprised of a cluster of various ethnic groups, many

which belong to the Bantu family (Worldometer 2017).

Its climate is mostly equatorial and rainfall and temperatures can vary drastically by region.

Southern and south-west Uganda generally experience consistent rainfall throughout the year,

while the north-east is dry and drought prone (Haggblade 2010). Rainfall averages in the

southern regions of the country are 1500 mm per annum and less than 500 mm in the

northeast. Reports state erratic rainfall is believed to be associated with climate change and

making the distribution of rainfall more volatile and uneven, with some heavy rainfall events

(NEMA 2010). Uganda’s climate and agro-ecological features are favorable for agriculture

and has been called a high-performing region despite its regional susceptibility and proneness

to drought (Masih et al. 2014; Leliveld et al. 2013). A large area of Uganda consists of high

elevation plateau, ranging from 1000 and 2500 meters above sea level (Ronner and Giller

2013). Figures 2.1 below depicts the average annual rainfall in millimeters and population

density across Uganda.


11

Figure 2.1: Rainfall (mm) and Population Density (Km Squared)

Source: NASA, 2000

The northern regions of the country experience unimodal rainfall and annual crops such as

maize, sorghum, groundnuts and sesame dominate. In the southern parts where it is bimodal,

the most common are perennial crops which include banana and coffee (Mubiru et al. 2012).

Uganda’s climate and agroecological zones are conducive to horticulture and has had much

success with cash crops, such as coffee, cocoa, tea, and flowers among others.

Most of its population coincides with areas of greater higher annual rainfall averages. The

central region of Uganda contains approximately 50% of its total population, nearing Lake

Victoria, as can be seen in 2.1.

2.2 Historical and Economic Context

Historically, there has been significant competition amongst various ethnic clans, much of

which has been over pasture and scarce natural resources such as water, livestock, and land.

Livestock rearing and its accumulation have significant cultural, economic and symbolic ties

in Uganda. Cattle husbandry is customary among men, and thus pastoral lands are of

significant cultural importance. Competition for land, water, forest products and mineral

resources have been a persistent trigger for inter-community and ethnic violence (Minority

Rights Organization 2011).

Uganda’s economic modernization and development has been accompanied by a series of

civil conflicts following independence. Economically, its history has been bleak, plagued by


12

cycles of famine until the early 1990’s, and has experienced less frequent, but still recurrent

food insecurity even until today. Recently, Uganda has become more peaceful and

increasingly stable with significant overall progress socially and economically (Minority

Rights Organization 2011). Table 2.1 provides a small summary of some of Uganda’s social

and economic statistics.

Table 2.1: Economic Overview for Uganda: 2017

Climate Tropical; semi-arid in northeast

Population: 41,699,654 (Worldometer 2017)

Age Structure:

0-14 years: 48.26% (male 9,223,926/female 9,268,714)



55-64 years: 2.5% (male 460,000/female 496,399)

65 years and over: 2.01% (male 337,787/female 431,430) (2016 est.)

Urbanization

Urban population: 16.1% of total population (2015)

Fertility

Total Fertility Rate 5.8 children born/woman (2016 est.)

Sanitation Access

(Improved)

Urban: 28.5% of population

Rural: 17.3% of population

Total: 19.1% of population

(Unimproved)

Urban: 71.5% of population

Rural: 82.7% of population

Total: 80.9% of population (2015 est.)

Literacy

Definition: age 15 and over can read and write

Total population: 78.40%

Male: 85.30%

Female: 71.5% (2015 est.)

Demographics

Population Growth Rate: 3.3% annually (World Bank)

Life Expectancy: 59.18 (World Bank)

Economy

GDP Growth Rate: 4.9% (2016 est.)

Agriculture: 24.50%

Industry: 21%

Services: 54.4% (2016 est.)

GDP Purchasing Power

Parity: $84.93 Billion (2016 est.)

GDP per capita: $2,100 (2016 est.)

Sources: CIA World Factbook, Worldometer, World Bank

With regards to agriculture, Uganda presents a fascinating case for study because it has

abundant natural resources. However, the country is faced with various problems that are


13

hindering the growth of the agricultural sector such as declining soil fertility, drought and

lack of proper investment in key areas that promote agriculture (FAOstat 2017).

Like many other African countries, agriculture is the primary sector of the economy.

Numerous sources have identified Uganda as having a wealth of natural resources and

agricultural potential. Up to 80% of Uganda’s land is arable, although it is estimated that

only 35% is being cultivated (FAOstat 2017). The share of agricultural production in total

GDP has declined over the past years, while its growth in the sector has remained around

2.6% per annum in the past eight years (World Bank 2016). Agricultural growth has been

largely hindered by supply-side constraints such as inadequate investments in a sector that

continues to depend on low levels of technology. There exists enormous potential in agro-

processing, although it has been relatively untapped (World Bank 2016). The sector still

comprises 24.5% of the nation’s GDP in 2015 (World Bank, 2017), employs approximately

80% of the labor force in 2016 (FAO 2016) and accounts for 85% of export revenues (World

Bank, 2017; USAID, 2016). Additionally, in 2010 agriculture contributed to approximately

40% of the manufacturing sector via food processing (NEMA 2010).

In Uganda’s colonial era, the British did not develop large-scale plantations with the

exception of tea and sugar estates, and later implemented cotton and coffee as a “forced

system of cultivation” (Leliveld et al. 2013: 422-3). Production of cotton and tea practically

collapsed during the 1970’s. The government attempted to promote diversification in

commercial agriculture that would broaden a variety of non-traditional exports (Chauvin et

al. 2017). Coffee became the primary export crop during the late 1980’s and has remained

until recently despite its decline of up to 17.9% of total export earnings. Data from FAO

illustrates Uganda’s agricultural productivity from 1961 to 2011 in figure 2.2. Much of the

productivity gains have been achieved through area expansion while lesser improvements

from yields, especially since the 1980’s. Over the time period, agricultural production has

not kept up with population growth. Population growth is over five times in 2013 of what it

was in 1961; area harvested grew approximately 50% over the time period; and yields have

stagnated since the 1980’s.


14

Figure 2.2: Uganda Production, Yields and Population Trends

Source: FAOSTAT 2017. Data includes all reported crops

Ugandan agricultural commercialization has been a key policy topic to break the stagnation

in agricultural productivity. Broad economic growth in Uganda is envisioned to take place

with large improvements in the agricultural sector, since other economic sectors are still in

phases of low development (Dorosh and Thurlow 2009).

2.3 Smallholder Farming

The majority of agricultural output comes from 3 million smallholding subsistence farms

with an average plot size of 2.5 hectares (FAO 2017). These smallholders constitute

approximately 85% of the people engaged in agriculture, while medium scale and large scale

constitute 12% and 3% respectively (DRT 2012). Smallholder farmers, whose definition is

usually of a farmer with up to 10 hectares of land (FAO 2016) own approximately 92.1% of

agricultural land in Uganda.

Table 2.2 provides a breakdown of the percentage of agricultural land under each landholding

category, as well as the average number of crops cultivated on that type of farm size. Data is

from the World Bank’s Living Standards Livelihood Survey (LSMS) in conjunction with the

Ugandan National Household Survey (UNHS).

0

5000

10000

15000

20000

25000

30000

35000

40000

0

50000

100000

150000

200000

2500001

96

1

19

63

19

65

19

67

19

69

19

71

19

73

19

75

19

77

19

79

19

81

19

83

19

85

19

87

19

89

19

91

19

93

19

95

19

97

19

99

20

01

20

03

20

05

20

07

20

09

20

11

20

13

Po

pu

lati

on

(in

th

ou

san

ds)

Hec

tare

s an

d H

ecto

gram

s/H

ecta

re

Year

Area Harvested Yields Population (in thousands)


15

Table 2.2: Share of Agricultural Land Owned by Farm-holding Size

Landholdings Percentage of Land Average Number of Crops Cultivated

Less than 1 Ha 24.65% 3.59

Between 1 and 5 Ha 56.75% 4.78

Between5 and 10 Ha 10.70% 5.33

Greater than 10 Ha 7.90% 5.57

Source: Uganda World Bank LSMS survey 2010/2011 and 2013/2014 n= 2,032

Farming in Uganda is predominantly rain-fed with use of low-cost inputs and is labor

intensive. Farming systems can generally be characterized as mixed and intercropped. Many

households combine plantains with cassava, millet, sorghum, sweet potatoes, beans,

groundnuts or maize, sometimes including a cash crop such as coffee, cocoa or tobacco

(Komarek 2010).

Infrastructure has remained undeveloped, especially in rural Uganda due to low investment

levels compared to other countries in the region (World Bank 2012). Government

expenditures in the agricultural sector were only 4-5% of the national budget from 2001-

2008, despite intra-regional pledges in the Malabo Declaration to commit 10% (AUC, 2014).

This has led to high input costs, low farm-gate prices, significant transaction costs and low

levels of commercialization (World Bank 2012). Missing or underdeveloped food markets

and perceived high uncertainty around price and yield are common disincentives to increase

production (World Bank 2012). Considering relatively high prices of inputs, scarcity of food

or insurance markets (absence and, or access), and yield risks, production for self-sufficiency

takes priority for most households. Where markets are more developed and accessible

allowing for higher and less volatile prices of non-staple crops, farmers in Uganda are more

inclined to specialize in higher-valued crops. According to the World Bank, the bottom 25%

of commercialized households only sell 4% of their agricultural produce in the market while

the top 25% sell more than 50% of their total production.

2.3.1 Diet and Consumption

Ugandan household diets are largely dependent on staple commodities. Table 2.3 below

illustrates the typical Ugandan diet. Plantains (cooking bananas), cassava, maize, sweet

potatoes and beans dominate the consumption categories, although pulses, nuts and green


16

vegetables also compliment the diet in smaller portions. Plantains, also called “matoke” or

“matooke”, have a distinct cultural connection to Uganda and plantains are considered the

primary staple (Haggblade & Dewina 2010). The main staple crops are broken down into

several consumption categories, where plantains, cassava and maize represent the bulk of

caloric intake. Other sources have identified plantains and beans as the most important in

consumption rankings, likely for nutritional reasons, despite beans ranking sixth in table 2.3

(Mulunba et al. 2012).

Table 2.3: Staple Food Consumption in Uganda

Commodity Annual Quantity consumed Daily caloric intake Calorie share In

Total Diet kg/capita kcal/day (percent)

Plantains 172 419 18%

Cassava 101 300 13%

Maize 31 266 11%

Sweet Potatoes 82 215 9%

Beans 16 148 6%

Wheat 7 42 2%

Rice 4 53 2%

Other 1133 48%

Total 2,360 100%

Source: Haggblade & Dewina, 2010

Three of the most important staple crops, plantains, cassava and sweet potatoes are

relatively untraded in the export market. However, maize and beans are widely traded.

Farmers usually produce a surplus of these crops which are traded within the region, mostly

to Kenya who experiences consistent maize deficits, along with other countries in the EAC

(Haggblade and Dewina 2010).

2.3.2 Agricultural Diversity

Various policies in the past two decades have promoted the commercialization of smallholder

farms. With mixed results, success has mainly been seen in increased export for products

that have values greater export values than $1,000 per ton. These products have been

predominantly coffee, tea, cotton, flowers, and fish (World Bank 2012). Their value can

sufficiently exceed farm-gate prices to make trading profitable despite high transactions

costs. There has been lesser success for products with export prices below $1,000 per ton.

Uganda’s top agricultural export products are coffee, raw tobacco, tea, maize and palm oil,

with much of its market being Europe and neighboring African countries (OECD 2016).

Since the 1980’s, coffee has dominated the export base in Uganda. Despite it still being its


17

primary export crop, there has been great diversification away from coffee in its export

structure (OECD 2016). In the 1990’s Uganda was one of the world’s largest producers of

coffee, which was an enormous contributor to GDP through exports, and played a significant

role in rural incomes. Since the early 2000’s, export prices for coffee dropped almost 70%

and it became apparent that new export products would be needed to supplant coffee. Crop

diversification to broaden the export market has since been a key objective, making Uganda a

relevant case study. Uganda’s mountainous landscape with high elevations can make

diversification into some crops other than coffee a challenge. Even though Uganda remains a

large coffee producer, other export products have begun to take its place, such as vanilla, tea,

spices, fish and horticulture (Dunkley 2017).

To demonstrate the uptick in the production of new crops, figure 2.3 illustrates the Shannon

Diversity Index (SDI). The SDI is an index commonly used to characterize the number of

species within an area, it accounts for both abundance and equitable composition of the crops

considered. In this case, we consider all crop species in aggregate for Uganda from 1961-

2013. The SDI displays an approximately 8% increase in diversity over the past 52 years. A

noticeable increase in the diversity score is apparent after 1990 when policies that promoted

crop diversification and commercialization were implemented.

Figure 2.3: Shannon Diversity Index - Uganda

Source: Bureau for Food and Agricultural Policy (2016).

Contrasting Uganda with other countries within sub-Saharan Africa as seen in figure 2.4,

Uganda exhibits a similar upward trend in diversity over the 52 year time span. Crop

diversity may be associated with the transition from subsistence to commercialized farming,

as smallholders could be shifting away from staple to higher valued crops (Africa


18

Agricultural Status Report (2016). Since many countries in SSA share the similar

development challenges, insights to Uganda could be transferrable to other SSA countries

with respect to factors that encourage crop diversity which could lead to increased

commercialization.

Figure 2.4 Shannon Diversity Index – Select SSA Countries

Source: Africa Agricultural Status Report (2016)

Despite a gradual increase in crop diversity as indicated in the SDI among these various

countries, impediments to smallholder crop diversification still remain. Many of the common

challenges are lack of high-quality packaging, lack of storage facilities, high freight costs; a

lack of road infrastructure in rural areas; a complicated and inefficient land tenure system;

and a lack of specialized skills (NAADS, 2016; Export Gov, 2016). Proper infrastructural

facilities that reduce transaction costs and facilitate market access are a key focal point to

induce commercialization and diversification among smallholder farmers (Jaleta et al. 2009).

In Uganda’s case, specialization is associated with areas of high levels of price and yield

certainty (World Bank 2012). Most Ugandans are considered relatively diversified to account

for such uncertainties. To demonstrate common levels of diversity among Ugandan farms,

the Herfindahl-Hirschman index is another widely used tool to measure land concentration.


19

The index rises asymptotically toward one as a farmer’s crop portfolio becomes increasingly

specialized and toward zero with greater levels of crop diversity. The Herfindahl-Hirschman

distribution for 75% of Ugandan farmers is below 0.27, which can be associated with

approximately three different crops in their mix (World Bank 2012). Table 2.4 provides a

breakdown of the household mean share of farmland allocated to the most common crops

from 2009/2010 agricultural season to 2013/2014 to demonstrate typical smallholder land

use.

Table 2.4: Household Mean Share of Agricultural Land Disaggregated by Crop

Crop Land allocation

2009/2010 2013/2014

Cassava 16.95% 16.56%

Maize 15.89% 13.78%

Beans 11.40% 16.46%

Sweet potatoes 8.17% 6.31%

Banana food 7.12% 12.57%

Groundnuts 6.24% 6.47%

Sorghum 6.14% 3.19%

Coffee all 4.48% 0.70%

Finger millet 2.99% 2.50%

Sugarcane 1.70% 0.54%

Pigeon peas 1.56% 1.83%

Field peas 1.50% 0.39%

Simsim 1.19% 0.80%

Rice 1.10% 0.57%

Tobacco 1.09% 0.25%

Banana beer 0.84% 0.98%

Irish potatoes 0.51% 1.50%

Sunflower 0.41% 0.33%

Soya beans 0.37% 1.01%

Horticulture 3.93% 4.12% Source: Uganda World Bank LSMS survey 2009/2010 and 2013/2014 n= 2,032

2.3.3 Farming Systems

Cassava and sweet potatoes are argued to be the most important crops of Uganda because of

their importance for food security (Ronner and Giller 2013). Next in importance are plantains

(cooking bananas), followed by coffee, maize and beans. The primary cereals produced are

maize, finger millet, sorghum, rice, pearl millet and wheat (Kabeere and Wulff 2008). The

major cash crops are coffee, cotton and tea.


20

Fermont et al. (2008) describes Uganda’s farming systems as dynamic and constantly

changing their strategies in response to increasing population densities and political and

economic variables (Ronner and Giller 2013). For example, in parts of Uganda, cassava has

frequently replaced cotton and millet as a method to reduce the negative impacts of soil

mining. Or, fallows are usually rotated as part of a farming system, although with increasing

population pressure, cassava is used an as “imitation fallow” in the case of poor fertility of

soils to replenish nutrients (Ronner and Giller 2013). Table 2.5 illustrates the percentage

share of farmers cultivating Uganda’s primary crops, as well as the most frequently occurring

crop portfolios from the 2010-2011 LSMS survey sample.

Table 2.5: Ugandan Crop Portfolios: 2010/2011

Crop

Percentage of

Households

Cultivating

Common Uganda Crop

Portfolios

Percentage of

Households

Cultivating

Cassava 66.5% Maize and Cassava 40.8%

Beans 61.1% Maize and Beans 40.3%

Maize 60.0% Cassava and Plantains 29.5%

Plantains 47.0% Maize and Plantains 27.3%

Sweet Potatoes 44.6% Maize and Sweet Potatoes 27.2%

Coffee 25.7% Maize, Cassava, Beans 26.6%

Groundnuts 24.6% Plantains and Coffee 23.0%

Sorghum 17.5% Maize, Sweet Potato, Beans 19.3%

Horticulture 16.5% Cassava, Coffee, Plantains 14.7%

Millet 12.0% Maize, Plantains, Coffee 14.3%

Peas 8.7% Maize, Coffee, Beans 12.9%

Irish Potatoes 4.0% Maize, Plantains, Coffee, Beans 11.5%

Sesame 3.9% Maize, Sorghum, Millet 3.0%

Cotton 1.8% Maize and Tobacco 1.1%

Tobacco 1.5% Cassava, Sorghum Tobacco 0.3%

Source: World Bank LSMS survey 2010/2011. Author’s own calculations n = 2,126

Fresh fruits and vegetables are non-traditional export crops. In the 1980’s horticulture

production began to gain traction resulting from economic strategies to broaden the export

base beyond its traditional exports coffee, cotton, tobacco and tea. Other non-traditional

products beyond horticulture that have contributed to Uganda’s growing export portfolio

mainly include fish/fish products, floriculture, horticulture, spices, hides and skins, and honey

(UNEP 2017).

Uganda is the second largest producer of horticultural products in sub-Saharan Africa after

Nigeria. Its biggest export market is the European Union, primarily Belgium and the


21

Netherlands. 40% of horticultural production is from smallholder farmers indicating there is

great potential to increase production.

Despite low intra-regional levels of trade for HVAs, increasing global consumption of these

products has led to new marketing opportunities for producers. From 1961 to 2007, area

harvested for fruits has increased 172.14% from 672,200 hectares to 1,829,370 hectares,

however from 2007 to 2015, the area decreased 45.87% to 991,790 (FAOSTAT 2018).

Vegetables have increased 10.29% from 54,907 hectares in 1961 to 60,562 hectares in 2015

(FAOSTAT 2017). These trends indicated that these products becoming gradually more

prominent as a sub-sector in agriculture. This presents an opportunity for increased

smallholder involvement.


22

CHAPTER 3: THEORETICAL FRAMEWORK

A large body of research exists on different aspects of cropping systems and land use. Over

the past 50 years, explanations of changing cropping patterns have emerged from varying

theories. Early studies largely examine land productivity and its association with various

dimensions of agricultural transformation on a macro scale. Later, much literature built on the

theories of the individual using the agricultural household model. The following section

provides a review of literature on crop diversification and its determinants, ending with

literature of the agricultural household model. The literature cited around crop diversification

is meant to provide an overview of not only its broad economic implications, but also how

farmers make decisions on which crops to cultivate based on a set of factors supported by

literature.

3.1 Crop Diversification

The degree of crop diversification can range from total specialization, to high levels of

diversity in which there are a greater variety of crops under cultivation (Turner 2014).

Regionally, crop diversification is based on its changing social, economic, technological,

geographical and institutional structure (Kumar 2004). From a smallholder’s perspective,

diversification decisions are made jointly bearing production and consumption decisions in

mind, while household, farm and regional characteristics are the main factors that affect their

land allocation decisions. Inderjit Singh (1985) proposes the canonical agricultural

household model that elucidates land allocation decisions. Much research on crop

diversification emerged following the Green Revolution that took place in Asia, such as

Minot2 (2006) and Ghosh3 (2014). More recently, studies that focus on these patterns in

Africa are becoming prominent.

3.1.1. Crop Specialization

Monoculture is the complete form of specialization, typically of grain crops in many

developing countries. Nevertheless, specialization can also refer to high levels of crop

2 Minot, Nicholas, ed. Income diversification and poverty in the Northern Uplands of Vietnam. Vol.

145. Intl Food Policy Res Inst, 2006. 3 Ghosh, Madhusudan, Debashis Sarkar, and Bidhan Chandra Roy, eds. Diversification of Agriculture

in Eastern India. Springer, 2014.


23

concentration. Kurosaki (2003) explains that the patterns of crop specialization and

diversification allow us to characterize the nature of that region’s agricultural transformation.

Recent literature has identified a U-shaped relationship between a nation’s level of crop

diversity and the extent of the market (Shilpi and Emran 2012). Lesser developed markets

often have higher levels of crop specialization, and as its markets develop, levels of crop

diversity increase before reaching economies of scale, returning to specialization.

Specialization occurring in economies of scale can be seen in the corn belts of the United

States, for example (Haplin 2000).

In lesser developed countries, specialization is often associated with low levels of

commercialization and rural market development (Haplin 2000; Kurosaki 2003; Minot 2006).

In Uganda, for example, rural areas exhibit many of the characteristics of a lesser developed

country, such as low density and quality of infrastructure, unsophisticated markets, and low

levels of technology. In such instances, where markets may be difficult to access due to high

transaction costs, production decisions are much related to food access. Therefore, the choice

and extent for a household to diversify its crop portfolio will depend on its level of

commercialization, whether it be subsistence, semi-commercial, or a fully commercial system

(Ahmadzai 2017).

Kim (1981) provides several detailed motives as to why a producer might choose to

specialize. The first deals with the availability of factor resources, which can range from land

typology, soil type, climate, to human resources such as crop specific knowledge and

expertise (Kim et al. 1981). Cultivating crops that are unfamiliar to farmers presents a risk,

so areas that have historical trends of monoculture or little diversity are difficult patterns to

break if proper expertise and working knowledge is not available (Minot 2006). Furthermore,

many developing countries lack diversified markets which hinders the ability to diversify. In

such instances, those countries are also frequently subsidizing a narrow range of commodities

which will incentivize further production of those crops (Haplin 2000). Generally,

constrained market access will hinder the range of products produced by limiting exposure to

information and access to inputs, increasing the prospect of specialization. Many of the same

dynamics which influence specialization are inversely related to diversification.


24

3.1.2. Crop Diversification and Determinants

In developing countries, agricultural diversification traditionally refers to a subsistence form

of farming where farmers are cultivating varieties of crops on a plot of land and engaging in

several enterprises on their farm portfolios (Vyas 1996). Common definitions hold that there

is a shift from less profitable to more profitable crops (Vyas 1996). Higher valued crops

usually include horticulture, spices, oilseeds and cash crops. The emergence of crop

diversification is an indication that two things are happening 1) changing business activities

where farmers are responding to changing opportunities such as new production technologies

and price signals in the market (Abro 2012) and 2) more efficient allocation of resources

(Mukherjee 2010). This results in greater economic returns, bringing about land productivity

increases with greater inclusion of high-value crops (Rao 2006).

The process of agricultural diversification resembles similar patterns of diversification in

non-agricultural sectors. In theory, the utilization of surplus capacity will lead to a related

firm’s diversification bound by the absence of market failure (Wernefelt and Chaterjee, 1991)

(Haplin 2000). The initial stage of diversification in agriculture can be viewed as a shift

away from monoculture (Haplin 2000). The resultant stage can be classified as when a farmer

has more than one crop enterprise and he can sell or produce as he chooses throughout the

year, known as the mixed farming stage (Metcalf, 1969; Shucksmith et al. 1989). Lastly, the

process of diversification transitions into activities beyond primary agriculture (Newby

1988). These activities often include on-farm processing and the creation of non-agricultural

products and services, as well as venturing into off-farm employment – a key step in

agricultural transformation (Evans and Ilbery, 1993; Reardon and Timmer 2014).

Changes in aggregate land productivity are associated with the efficiency of resource

allocation (Kurosaki 2003). Land use is a substantial part of resource allocation decisions,

which includes what crops to plant and what proportion. In theory, a nation will use its

resources optimally and will therefore allocate its land to maximize output. From a policy

point of view, its promotion is meant to increase the country’s self-sufficiency and provide a

broader export base (Abro 2012). At the farm-level, it is becoming widely recognized as an

avenue to increase and stabilize farm incomes (Mofya-Mukuka et al. 2016). Despite there

being many individual determinants that are driving or hindering a farm’s level of diversity,


25

they can be broadly summarized within the contexts of risk, infrastructure and market

development (Dutta 2011).

Risk

Risk and uncertainty are inherent features in agriculture (Ali 2015). Income uncertainty can

lead to substantial welfare costs for households. Small-scale farmers, especially in developing

countries can be unfavorably affected by drought or price fluctuations and diversification can

be used as a strategy to hedge against such risks and uncertainty (Minot 2006). Crop portfolio

diversification is analogous to the diversification of marketable stocks or bonds and these

decisions are made to minimize risk and maximize long term gains (Moore and Snyder

1969). Small-scale farmers are believed to exhibit risk-averse behavior which is associated

with the decision of which crops to plant and how much land to allocate to them (Fafchamps

1992). The primary factors that contribute to perceived farming risk are prices and yields,

which are affected by a host of factors (Dercon 1996).

Mcguire (1980) conducted a study on production patterns across various states in the United

States analyzing southern farmers who cultivated corn and cotton. His research aimed to

rectify previous controversy among economists who theorized that southern farmers were in

fact risk ‘gamblers’ rather than risk-averse. This initial claim was in response to an upsurge

in the proportion of land devoted to cotton production away from maize because cotton was

perceived as a riskier crop due to its market price volatility. His analysis sought to understand

if farmers selected their crop portfolios in a risk-averse or risk-seeking manner, based on

expected prices. A behavioral model that calculates farmer’s subjective risk coefficients in

relation to land allocation decisions was employed. Of his sample of 45 farmers over eight

years, all of them displayed positive coefficients indicating that farmer’s attitudes toward risk

were risk-averse and homogenous despite heterogeneous populations.

Empirical approaches have been taken to quantify the exposure of downside risk under the

production of various crops. Di Falco and Chavas (1996) used a stochastic production

function to assess welfare effects from biodiversity on production risk. Since farmers are

considered to be risk averse, he hypothesizes that risk exposure will make farmers worse off,

which implies a positive cost of risk. Avoiding risk exposure indicates that there will be an

incentive to grow a variety of crop cultivars that will increase the skewness of the distribution


26

of returns. His findings indicate that biodiversity is strongly related to the skewness of

production. The results concluded that land fertility and crop biodiversity were substitutes –

agreeing with theory. This implies that greater number of crop varieties reduce the exposure

to downside risk in areas where crop failure is more likely.

Infrastructure

Inadequate transport and communications infrastructure in remote areas has been obstacle to

improving productivity, income, and food security in rural Africa (Turner 2014). Farmers

have been impeded by these constraints when trying to purchase inputs or sell their surplus

(Barrett 2017). In many lesser developed countries, markets for high-valued agricultural

products are concentrated in urban and semi-urban areas (Rao 2006). Transport

arrangements from rural to urban areas are often scant which results in increased transaction

costs for producers, discouraging market participation and thus, diversification toward

higher-value products. Access to markets, transportation facilities and post-harvest

infrastructure have been cited as critical aspects to the growth of high-value products and

crop diversification, although infrastructure has been considered to include many other

aspects such as roads, electricity, telecommunications, transport, and irrigation facilities (Rao

2006).

Numerous studies have examined how different infrastructural facilities play a role in

agricultural household decisions. Rao (2006) studied the impact of roads and urbanization on

the production of high-value commodities in India from 1980 to 1988. Road and population

density are his key variables of interest in relationship to regions of high-intensity production

of high-valued commodities. His findings show that greater land concentrations of high-

valued commodities were in regions with greater population pressure, higher road density,

and a larger urban population. Concentrations of high-valued commodities decrease with

distance to urban centers. High value agriculture comprised 4 percent of the value of

agricultural output in urban areas while only 32 percent in far urban areas. He argues this

occurrence suggests urbanization is an important factor in the growth of high-value

production from the demand side, while greater supplies of labor and infrastructure facilitate

the process.

a


27

Markets

Abounding literature highlights the negative impacts of market failure and how agricultural

productivity cannot thrive in its presence. From a micro or macro perspective, market

development has been widely cited as a key force in driving structural transformation, as well

(Timmer 1988; Barrett 2017). Market development facilitates crop diversity by building

working capacity, connecting farmers with information and inputs.

Early attempts to properly describe peasant farmer’s behavior were confounding and many

researchers believed the utility maximization approach was an ineffective technique proposed

by formal economists. Governments were frustrated by policies that were rendered

ineffective which aimed to promote technology adoption and price stability of crops because

they saw no change in farmer’s behavior. De Janvry (1991) offered a seminal explanation of

market failure from the perspective of the agricultural farmer to remedy the debate. De

Janvry described how previous theories did not adequately take into consideration all of the

peasant farmer’s choice variables in maximizing utility. In other words, the incentives

thought to be in place to elicit a reaction from farmers were constrained by unseen

information, such as high transaction costs when selling their produce.

Immink and Alarcon (1993) examined the complete substitution between food and cash crops

to explore changing patterns in farmer’s crop mixes during a period of commercialization of

small-scale farmers in Guatemala. With nationally representative cross-sectional data, they

categorized households into four groups of different crop mixtures and estimated the

probability of a household choosing one of the crop portfolios. Their key finding was that the

lack of access to credit markets was a key constraint to technology adoption.

Various authors support the findings that access to inputs are essential to agricultural

productivity. Chavas (2001) conducted a study using nonparametric methods estimating a

variety of productivity indices. The analysis studies twelve developing countries between

1960 and 1994 to analyze the contribution of each type of input to aggregate production. The

time-series results disaggregated the factors associated with agricultural output. Technology

remained relatively the same over this time period and the greatest contributor to output was

inputs such as fertilizer and pesticides.


28

Kibaara et al. (2009) examined the drivers of agricultural productivity using household panel

data in Kenya based on the availability of input markets. The study analyzed output changes

in maize, tea, coffee, sugarcane, cabbages, Irish potatoes and dairy. A Cobb-Douglas

production function was used to measure productivity changes. Increased productivity in

maize was the result of greater numbers of small holder farmers using fertilizer, improved

seed varieties, and the availability of fertilizer retail outlets. Increased dairy output was

described as the largest response to increased investment, indicating access to financial

services as the key driver.

3.2 Agricultural household models

The theory of the agricultural household model has been the building blocks for many

researcher’s framework in agricultural economics. Originally created as a method for price

policy analysis, it has been applied to a wide range of topics from technology adoption,

production functions, deforestation, to biodiversity. Since its origins, dating back to

Chayanov (1925), nuanced versions have come to incorporate various features seen in lesser

developed countries’ rural economies, such as imperfect or missing markets for inputs and

outputs, labor and the existence of transaction costs, among others (Taylor et al. 2002).

Currently, the framework of Singh (1985; 1986) has emerged as some of the most recognized

and replicated. The following sheds light on the origins, evolution and theoretical framework

of agricultural household models and how it is relevant to this study.

Chayanov (1925) created one of the first agricultural household models of resource allocation

based in rural Russia. It sought to explain the role of the rural peasant economy as the country

transitioned from feudalism to socialism. Agricultural productivity was analyzed, but limited

to the nuclear household, thus making own demographic characteristics explain to a large

extent its outcomes (Hammel 2005). Its formulation, therefore only included a product

market, but no labor market. Its main criticisms and limitations have been his exclusion of

factors beyond agricultural, such as labor and capital flows (Shanin 1986).

Mellor (1963) and Sen (1966) among several others,4 later formulated their own agricultural

household models that built on Chayonov’s earlier principles. Their work added by exploring

output outcomes considering variable inputs and the demand for leisure. Through the income

4 Tanaka (1951), Nakajima (1957; 1969)


29

effect, households may respond negatively to output prices if they have a large demand for

leisure (Singh et al. 1986). Nakajima (1969) posited that labor supply and output price had a

substitutable effect. Together, the work of Nakajima and Sen showed that the marginal

utilities of income and leisure were relatively constant (Singh et al. 1986).

An updated portrayal of factor markets, specifically for hired waged-labor was provided by

Barnum and Squire (1980) to further explore the relationship of output and labor demand and

supply. The results of their models showed that labor demand will respond positively to

output price in the event that it is not an inferior input, once they extended crop production

beyond staples to include cash crops (Singh 1986). Jorgenson and Lau (1969) followed this

by developing a separable subsistence model that allowed for household labor to be marketed

and household production is only partly for own consumption (Singh 1986).

Singh built his own model that incorporates many of the foundations provided before him.

Singh’s form was developed as a tool for price policy analysis in rural Japan (Kuroda and

Yotopoulos, 1978) tailored to the circumstances of developing countries to reflect missing or

incomplete markets for outputs and inputs (Taylor et al., 2002). An increase in the price of a

staple crop did not elicit the expected response policy makers thought it would have on

farmer’s marketed surplus. In efforts to comprehend this behavior, a model was proposed

which simultaneously considered both production and consumption when making household

decisions.

Prior economic theory dealt with production and consumption decisions separately, but the

reality is that in developing countries, agricultural households often integrate their

consumption and production decisions (Singh, Ahn and Squire 1981). Such decisions are

integrated by the necessity to select inputs for household crop production, while consumption

decisions are made based on the allocation of farm income and production to purchase goods

and services. A household’s profits are implicitly determined by household production minus

consumption of its own goods, and purchased goods. Singh’s standard model consists of a

utility function defined by consumption for each household member with a budget constraint

that includes production on owned assets. Household utility is a function of the consumption

of goods and leisure time. Maximizing utility is with respect to consumption, leisure, hired

labor, land, household labor, rented land and used land on the farm. This is bound by the

budget constraint which includes cash expenditures, hired labor, and own rented land which

cannot be greater than its own cash revenue (Singh 1986).


30

Various studies have applied a similar framework to different areas within agricultural and

resource economics. This study extends the model to question which crops to produce and

how much land should be allocated to each. Econometrically, estimable versions of these

models have been applied to crop choice, land allocation and diversity outcomes (Benin

2003). The following will provide greater insight to the empirical structure of the agricultural

household model. Below follows Singh’s (1986) basic model framework, then provides the

multiple crop model to demonstrate crop choice and land allocation decisions. This will

serve as the foundational theory as to how households make diversification decisions toward

HVA.

3.2.1 Basic Model

The utility function can be illustrated as, max𝐴𝑎,𝑀𝑚,𝑋𝐿

𝑈(𝐴𝑎, 𝑀𝑚, 𝑋𝐿) where the objective is to

maximize expected utility from household-produced goods, purchased goods, and leisure

while facing several constraints. In the utility function, 𝐴𝑎 is production of an agricultural

staple, 𝑀𝑚 is a good that is purchased in the market, and 𝑋𝐿 represents leisure time in hours.

Under these conditions, the household faces the following constraints of; cash income, time,

and production technology. First, the cash income constraint can be defined as:

𝑝𝑚𝑀𝑚 = 𝑝𝑎(𝑄𝑎 − 𝐴𝑎) − 𝑤(𝐿 − 𝐹) − 𝑧𝑣𝑉 + 𝐸 (1)

Intuitively, this constraint implies that a household can purchase in the market only with the

earnings from its own surplus production. In the budget constraint equation, 𝑝𝑚 represents

market purchased goods while 𝑝𝑎 represents the price of the household producing its own

agricultural products; 𝑄 stands for the quantity of agricultural products produced by the

household such that 𝑄 − 𝐴𝑎 is the household’s marketed surplus. The wage rate is denoted

as 𝑤 in the market and 𝐿 and 𝐹 are total labor inputs and family labor inputs, respectfully.

When 𝐿 − 𝐹 is positive it indicates the household hired labor and when negative, the

household has off-farm labor supply. Variable inputs such as, fertilizer or seed are

represented as 𝑉, while 𝑧𝑣 is its market price. Household income, 𝐸 is earned not from labor

or from the farm.

Secondly, the time constraint can be illustrated as:

𝑋𝐿 + 𝐹 = 𝑇 (2)


31

Here, leisure time plus on farm or off-farm employment cannot exceed the total available

time to the household, 𝑇. Lastly, the production constraint that is thought to be based on the

household’s production technologies illustrates the relationship between inputs and output as:

𝑄𝑎 = 𝑄(𝐿, 𝑉, 𝑍, 𝐾) (3)

The new variables presented in this constraint include 𝑍, the household’s fixed quantity of

land and 𝐾 represents the household’s fixed level of capital. Ultimately, this constraint

describes how production is a function of labor inputs, variable inputs, and fixed quantities of

land and capital.

The constraints can be then substituted into a single equation that will form part of the final

utility equation. To collapse all of the constraints into one equation by substituting where

necessary, we have:

𝑝𝑚𝑀𝑚 + 𝑝𝑎𝐴𝑎 + 𝑤𝑋𝐿 = 𝑤𝑇 + 𝜋 + 𝐸 (4)

The left-hand side of constraints equation can be seen as total household expenditure on

market-purchased goods, the household’s expenditure of its own output, and the purchase of

its own time in the form of leisure. The right-hand side of the equation (x) is Becker’s

development of the full-income concept which denotes the value of the stock of time, 𝑤𝑇

owned by the household, plus farm profit, 𝜋, plus off-farm income, 𝐸, is equal to total

household profits. With this information, the equation for farm profits can be derived as:

𝜋 = 𝑝𝑎𝑄𝑎(𝐿, 𝑉, 𝑍, 𝐾) − 𝑤𝐿 − 𝑧𝑣𝑉 (5)

The current model assumes that the household produces one crop, 𝑎. To analyze crop

diversification decisions based on the agricultural household model, we must further develop

the model to include multiple crops.

3.2.2 Multiple Crop Model

To understand how a household makes decisions on how to allocate its land between multiple

crop choices, I borrow the conceptual model from Turner (2014) and Benin et al. (2004) that

has been extended from Singh (1986). We assume the farm-household produces a vector of

outputs, 𝑄 from a vector of inputs, 𝑋. Building on the equations from the single crop model,

the farmer’s production function for each crop, 𝑗 becomes:


32

𝑄𝑗 = 𝑓(𝑿𝑗𝑘 , 𝛼𝑗|𝐴, Ω) (6)

𝐴 is a fixed amount of land for the household, as before. Household utility is derived from

different consumption bundles, and its levels depend on the preferences within the household.

We therefore assume that preferences are determined by characteristics of the household such

as, age, education and wealth, which will be represented as Ω; and 𝛼𝑗 represents the share of

land, 𝐴 allocated to crop, 𝑗. The profit equation can then be generated by the sum of outputs

for each crop minus the inputs costs which can be seen as:

𝜋 = ∑𝑗=1𝑗𝑝𝑗𝑄𝑗 − ∑𝑘=1

𝑗𝑤𝑘𝑋𝑗𝑘 (7)

Here, 𝑝𝑗 and 𝑤𝑘 represent vectors of output and input prices, respectively. The equation is

arranged to maximize the household’s expected utility:

𝐸𝑈 = (𝜋(𝑄, 𝑋, 𝑝, 𝑤|𝐴, Ω) (8)

And optimal level of inputs are then found by applying the first-order conditions with respect

to 𝑋, such that:

𝑋𝑗𝑘∗ = 𝑋𝑗𝑘

∗ (𝑝𝑗 , 𝑤𝑘 , 𝑈|𝐴, Ω) (9)

Finally, we can then solve for the optimal output level of each crop 𝑗 which is dependent on

𝑋𝑗𝑘∗ as:

𝑄𝑗∗ = 𝑓(𝑋𝑗1

∗ …𝑋𝑗𝑘∗ )|𝐴, Ω (10)

Transaction costs, rather than realized prices are represented in household characteristics like

physical access to markets, therefore are better represented in a new vector, 𝜗 for market

characteristics. However, Turner (2014) contends that regional market prices are an adequate

substitute because they observe prices at large, which will capture exogenous price variation

that is likely unaffected by the individual household. Under Turner’s modification to Benin,

we can account for district-level market prices as 𝑝 and transaction costs in the form of

physical market access that falls under market characteristics. The optimal crop mix

expressed as the optimal level of land 𝛼𝑗 , allocated to each crop 𝑗 so that, ∑𝑗𝐽𝛼𝑗 = 1, 𝑗 =


33

1,2, … 𝐽, is a function of farm size, market prices, market and household characteristics such

that:

𝛼𝑗∗ = 𝑓(𝐴, 𝑃, Ω, 𝜗) (11)

Broadly defined, these are the hypothesized factors that affect household crop diversity. The

applied econometric approach will attempt to identify the relevant variables that have been

mentioned in the agricultural household model holding that households behave to maximize

expected profits through agricultural production, and therefore having the optimal crop mix.

3.3 Previous Application of Crop Diversification Models

Application of fixed effects estimations have previously been used to examine a wide range

of crop diversification outcomes. Many fixed effects approaches when applied to crop

diversity analyze nutritional or crop yields and their relationship with crop diversity. More

recently, household models that explore the determinants of household crop diversification

have been undertaken. Kurosaki (2003) provides an analysis of the dynamics of crop

diversification at the district level to understand cropping patterns of subsistence farmers in

the Western Punjab with data from 1903-1992. This study utilized the Herfindahl-Hirschman

index to characterize temporal and spatial crop shifts. Results concluded that changes in

district-level crop diversity reflected that regions, comparative advantage is based on its

geographic features and level of market development.

A study conducted in India by Birthal (2007) explores the hypothesis that diversification

toward high-value crops, such as fruits and vegetables can reduce poverty through

smallholder participation with fixed effects. Population density, urbanization, and per capita

incomes are the main controls from the demand side. Water availability, production

technology resource endowment, and infrastructure features such as roads and markets are his

hypothesized supply side factors that facilitate diversification (Birthal 2007). He states that

diversification toward fruits and vegetables usually requires greater levels of capital, relative

to other crops. The key variables of interest are access to credit and capital related factors

since many higher valued products may be capital intensive. These variables proved to be

significant in crop diversity shifts toward high value products. Results also suggested that the

uptake of riskier crops was considered as deterrents to smallholder participation, while crops

such as fruits that required less prior working knowledge were more frequently adopted.


34

Abro (2012) used a generalized least squares model with fixed effects to examine the impact

of different forces on crop diversification in Pakistan using panel data from 1980-2011. Tube

wells for irrigation and length of roads were both significant variables which contributed

positively toward diversification. Tube wells provide water for irrigation. Thus, proving that

the availability of water is one of the determinants of crop diversity through high paying off-

season crops such as vegetables in Pakistan (Chand 1995).

In Zambia, a study was conducted by Mofya-Mukuka et al. (2016) to analyze the factors

influencing household crop diversity. The dependent variable used was the Simpson Index of

Diversity to express crop diversity with a fractional response probit and correlated random

effects for the model estimation as prescribed by Wooldridge (2010). The motive of the

study was to better understand why diversification beyond staples in Zambia is so little. The

key outcome of interest was to explore how government subsidies impact household crop

diversity. Results indicate that, since the government provides input and marketing subsidies

for maize, it negatively impacts smallholder crop diversification. Furthermore, extension

service contact, productive assets, hours to nearest urban center, annual average rainfall and

landholding size were all significant variables with positive relationships to household crop

diversity.

3.3.1 Research Application

The triple-hurdle model is a modern technique that was developed and first implemented by

Bill Burke in 2009. It has been developed to test its validity surrounding market participation

questions against commonly used approaches, such as ordered tobit models and double-

hurdle models. Its framework is built on previous double-hurdle models that consist of two-

stage decisions and outcomes. Traditional double-hurdle approaches to market participation

in agricultural economics are misspecified unless all of the population in the sample are

producers of the good in question, as with staple commodities. When studying a less

commonly produced good, such as dairy of HVA, this method is less appropriate. This

limitation can be overcome by including a third hurdle; to model the initial decision to

produce. This technique improves inference by addressing the fact that the decision to

produce or cultivate the good may or may not be driven by a different structural process than

participation in marketing (Burke et al. 2015). In real application, policies and other

incentives that induce market participation of HVA may also induce non-producers to being


35

producing, leading to significantly different estimates of an overall effect. This could provide

useful policy information.

The following will provide the intuition and rationale of the double-hurdle, a review of its

application and the framework laid out by Burke et al. (2015) for a triple-hurdle estimation

that will be used in this study.

3.3.2 Review of Double-Hurdle Applications and Triple Hurdle Background

A dilemma in econometric modelling can arise when an outcome of interest depends on a

prior condition being met that less frequently occurs within the sample population. Models

of agricultural market participation are prime examples of this problem. If a researcher

chooses to examine how smallholder farmers participate in markets and their quantity of

maize sold, the second outcome can only be realized if the household is a market participant.

Thus, in the event that a household does not participate, the value of sales will be zero. This

requires utilizing an approach where the parameters of the model can be estimated non-

linearly, making traditional ordinary least squares an inappropriate method for this research

question. Nuanced approaches to account for statistical challenges such as these use corner

solution models. The rationale for the double-hurdle method is to simultaneously analyze the

decision not to participate in the market which could provide valuable information, rather

than ignore those observations. Double-hurdle models have been developed to examine cases

where an event may or may not occur, though if it does, it takes on a continuous value (Burke

et al. 2015; Mather and Jayne 2011).

Foundations for double-hurdle models are built on Tobin’s (1958) tobit model. The tobit

model, was designed to estimate linear relationships between variables when there is

censoring of the dependent variable (Tobit 1958). Cragg (1971) developed a model as an

“alternative” to Tobin’s that has become widely used for its flexibility in comparison with

Tobin’s approach. Cragg’s model differs by breaking the tobit model into different

components where only the first stage parameters in the model determine the probability of a

limit observation, followed by a separate set of parameters to determine the density of non-

limit observations (Lin and Schmidt 2004). The rationale for a double-hurdle application is

to observe an outcome conditional on that outcome being realized.

Applications of double-hurdle models have become popular in agricultural economics,

specifically for studies dealing with market participation and a following continuous

outcome. Goetz (1992) was the first to borrow this framework in an application to market


36

participation and output quantities. In his two-tier approach, he first uses a probit to gauge

the probability that a household participates in the market via sales or purchases. The second

tier explores the quantity of how much is sold or bought in the market.

Barrett (2005; 2006) has employed double hurdle models on a variety of occasions.

Bellemare and Barrett (2006) evolve Goetz’s model to include an ordered probit in the first

stage to measure the probability of a household being net buyer, net seller, or autarkic. The

second outcomes uses a truncated normal regression to measure the quantities bought and

sold.

Turner (2014) uses correlated random effects with Cragg’s double-hurdle to estimate pigeon

pea cultivation and its area allocated with Mozambican panel data. She includes correlated

random effect techniques because she argues unobserved time-constant factors influence the

decision to cultivate pigeon peas. Thus, her first stage is the decision to cultivate pigeon pea,

and conditional on this outcome, how much area is then allocated to pigeon pea as a share of

total area. The key focus was to understand how infrastructure variables, particularly radio

ownership and mobile networks impacts on pigeon pea cultivation. Households who own

radios are believed to have greater access to information, in turn, their area allocated to

pigeon peas may be more responsive to information on prices. Her results found that among

these variables, radio ownership and pigeon pea price were significant in influencing the

decision in how much land is allocated to pigeon peas, if the household is a pigeon pea

producer.

More recently, Burke et al. (2015) extends the double-hurdle framework to include a third

hurdle. As has been seen in double-hurdle models, participation decisions unfold in two

steps; (1) an initial decision whether they participate and (2) what quantity to buy or sell.

Such models have been formulated around goods that are commonly produced, such as staple

crops. Therefore, this method is appropriate because typically all of the population in the

sample produces the good. However, the analysis would then change if it were extended to

include crops more seldom produced by smallholders, such as dairy, cash commodities, or

horticulture. Burke argues that for inference applicable to the broader population, the

decision to cultivate the crop in question should be the initial hurdle. Thus, a third hurdle

would be added to the existing double-hurdle model.

Burke applies this method to smallholder farmers in Kenya and their decisions on market

participation in the dairy sector. His model’s decision process becomes 1) Does the


37

household produce dairy 2) is the producing household a net buyer, autarkic, or a net seller,

3) what are the quantities bought and sold amongst market participants? Stage one represents

a binary decision of yes or no, and a probit model is employed. Stage two classifies

smallholders into three categories and makes use of an ordered probit. Lastly, two log-

normal regressions are used to analyze net volumes bought or sold (Burke et al. 2015). A

graphic illustration of the decision process can be seen in figure 4.1 as:

Figure 3.1: Triple-Hurdle Modeling Framework

Source: Burke, 2015

Burke’s aim was to analyze the factors associated with smallholder farmer’s participation in

dairy markets since it is considered to have great potential to enhance incomes. The findings

from his research show that a triple hurdle model can be superior to previous methods such as

double-hurdle or ordered tobit models. Burke’s results indicate that improved technologies

play a large role in the participation of dairy production and marketing. Furthermore, access

to land, the value of household productive assets, infrastructure and improved technologies

were statistically significant factors related to the probability of being a producer and being a

net seller. Infrastructure examined distance from electricity, which is statistically significant

and the probability of being a seller is inversely related to its distance. Considering that dairy

production can have significant initial investment, access to credit proved to be an important

factor to dairy market participation. One of the highlights of Burke’s findings show that

where dairy cooperatives are present households are 9-10% more likely to be net sellers,

unconditional on whether or not households are producers.


38

CHAPTER 4: EMPIRICAL FRAMEWORK

The objective of the study is to identify the factors that influence the cultivation of high-value

agricultural products in Uganda, household market participation outcomes, and net sales (for

net sellers) or net purchases (for net buyers) in the market, conditional on being HVA

producers. To address this objective, the triple-hurdle model is utilized. The empirical

specifications and methods of estimation are discussed in this chapter.

4.1 Methodology: Triple-Hurdle Model

The empirical model used to understand the factors that underpin a household’s decision to

produce and participate in markets of high-value products, relies on the economic

justification of the market participation models discussed in Chapter 3. Following Burke et

al. (2015), a triple-hurdle framework models the three stages of decision-making that

Ugandan HVA farmers face, these include;

1. Stage 1: The decision to produce high-value products (as defined in Chapter 2)

2. Stage 2: The decision to participate in the market as either a net buyer, net seller or

neither (autarky).

3. Stage 3: The degree of market participation, i.e. the net volume bought or sold

Figure 4.2 below illustrates the household decision tree.


39

Figure 4.1: Decision Pathways: Triple-hurdle Modeling Framework

To identify the key factors influencing the decision to cultivate HVA and household market

participation decisions that will be used in the triple-hurdle estimations, this study follows the

approach of Mofya-Ukuka et al. (2016) and Burke et al. (2015). The factors determining the

decision to cultivate HVA and participate in the market are hypothesized to be a function of

demographic, farm, and regional characteristics. These can be seen as;

Demographic characteristics include gender, age and level of education of the household

head, and full time adult equivalents. Female farmers generally face greater social barriers

than men (Farnworth, Akamandisa, and Hichaambwa 2011). Age of household head is a

proxy for experience which tends toward better farming practices, but may be less amenable

to cultivate non-traditional crops. Full time adult equivalents represent a form of labor

endowment and HVA can be capital intensive.

Farm-level characteristics include access to extension services, use of improved seed and

hired labor, value of household productive assets, and household landholding size. Elevation

of the plot, in meters above sea-level, access to extension services is a binary variable

indicating yes or no to household extension service contact, along with whether the

household used improved seed or hired labor. Value of household productive assets is a

proxy for wealth which is widely believed to lead to different farming outcomes (Carter

2000; Mofya-Mukuka et al. 2016). Household landholdings is thought to help explain


40

diversification decisions because households with less land may be less inclined to devote

land to non-consumption crops.

Regional characteristics include annual average rainfall, distance to nearest market,

population center, and nearest road, share of the district receiving credit, district prices for

Ugandan HVA substitutes such as maize, beans, cassava, as well as the price of an HVA –

Coffee. Infrastructure barriers such as distance to roads represent a form of transaction cost

because they limit a farmers’ access to distribute and sell their products. Long – term

average rainfall is an agro-ecological indicator that influences the comparative advantage of

producing staples.

The triple-hurdle model as proposed by Burke et al. (2015) uses full maximum likelihood to

estimate all of the parameters in each stage of the model. This will be discussed more in the

next section of this chapter. First, the justification of the selected methods and their

conceptual derivations are laid out which will provide the foundations for the likelihood

function to be estimated.

Stage 1. The Decision to Produce HVA

In stage 1, whether the household produces HVA? is either 1 (yes) or 2 (no). Unlike linear

models that primarily examine continuous outcomes, the interest in binary response models

lies in the response probability (Wooldridge 2013).

Linear probability models have been used to describe response probabilities for a sample

population. However, dependent variable outcomes behave non-linearly due to the

distribution of the marginal and are bound between the values of zero and one. In ordinary

least squares (OLS) an issue arises because when using linear predicted probabilities, values

can fall outside of this zero to one interval. Despite this problem, linear probability models

have been seen as good approximation of response probabilities, although other models have

been developed to that fit the non-linear nature of a sample population’s response probability.

Conventional approaches that solve this problem make use of probit and logit models to fit a

non-linear function to the data.

Probit and logit models transform the outcome of response probabilities by indexing the

explanatory variables into the cumulative density function over the interval zero to one.

Wooldridge (2013) refers to this as the index model which can be seen as;


41

𝑃(𝑦 = 1|𝑿) = Φ(𝑿𝜷) = 𝑝(𝑿) (12)

Where 𝜑 is the cumulative distribution function which takes a specific form depending on the

underlying economic model, which maps 𝑿𝜷 = 𝛽0 + 𝛽1𝑥1 + 𝛽2𝑥2 +⋯+ 𝛽𝑛𝑥𝑛 over the

response probability. This can also be represented as;

𝑃(𝑦 = 1|𝑿) = 𝑦 = Φ(𝛽0 + 𝛽1𝑥1 + 𝛽2𝑥2 +⋯+ 𝛽𝑛𝑥𝑛) (13)

Equations (12) and (13) can be interpreted as the probability of an event occurring

conditional on the exogenous explanatory variables (X) according to a vector of the

parameters to be estimated (𝜷). By making use of the Bernoulli formula, that predicts binary

outcomes, where in the event that a response does occur, is equal to the cumulative density

function (CDF) denoted as Φ, which is the covariates distribution, X, when y=1. The index

model uses a latent variable model to derive the CDF from the underlying model, which takes

the form;

𝑦 = 𝑿𝜷 + 𝑒, 𝑦 = 1 𝑖𝑓 𝑦 > 0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒, 𝑦 = 0 (14)

Where the event occurs, 𝑦 = 1 if [𝑦 > 0], e is a continuously distributed variable independent

of X symmetric around zero and Φ is the CDF of e thus describing how in equation (12)

that Φ(𝑿𝜷) = 𝑝(𝑿) (Wooldridge 2013).

Both logit and probit models make use of a CDF, but use different functional forms. The

logit assumes the CDF of the logistic distribution and the probit assumes the standard normal

cumulative density function. The probit’s interpretation involves marginal effects associated

with changing probabilities of the response outcome given a unit change in an explanatory

variable, ceteris paribus. Despite their differences, they usually yield similar results and their

preference is often discipline specific. In economics, probit models tend to be more intuitive

due to their marginal effect interpretation.

The interpretation of the parameters in a standard probit regression should be interpreted

based on its sign, while no specific inference can be made given the isolated effect of the

parameter on the dependent variable. Coefficients with positive signs indicate that increasing

values of a given variable’s coefficient will have greater probabilities of y=1, while


42

increasing values of variables with negative coefficients indicate decreasing probabilities of

y=1. The marginal effects in a probit model offer the isolated change in the response

probability given an incremental change in an individual explanatory variable, all else

constant. Building on equation (13), to obtain the marginal effects from a probit regression, it

requires taking the derivative of a CDF with respect to a given explanatory variable. This

will yield the marginal impact of on the response probability by changing an explanatory

while setting other explanatory variables to specified value. The marginal effect can be seen

as;

𝜕𝑦

𝜕𝑥𝑖= 𝛽𝑖Φ(𝛽0 + 𝛽1𝑥1 + 𝛽2𝑥2 +⋯+ 𝛽𝑛𝑥𝑛)

(15)

Probit models parameters are estimated via maximum likelihood. The following first

estimates the distribution of y which will be used in the log-likelihood function for each

observation i, however, the econometric estimation of the log-likelihood function will be

described in the next section. The following shows the distribution of 𝑦𝑖 given 𝑿𝒊 which can

be seen as;

𝑓(𝑦|𝑿𝒊, 𝜷) = [Φ(𝑿𝒊𝜷)]𝒚[1 − Φ(𝑿𝒊𝜷)]

𝟏−𝒚, 𝑦 = 0, 1

In the context of the triple-hurdle, the structure of the first stage decision, whether or not to

produce HVA can be represented with a probit model. For simplicity, the conceptual

framework will build on the following notation;

𝑐1𝑖 = 1[𝑦1 > 0] (16)

𝑐1𝑖 = 0[𝑦1 = 0] (17)

In the first stage the conceptual framework for the standard probit can then be seen as;

𝑃𝑟(𝐶𝑢𝑙𝑡𝑖𝑣𝑎𝑡𝑒 𝐻𝑉𝐴) = 𝑃𝑟(𝑐1𝑖 = 1|𝑿𝟏𝒊) = Φ(𝑿1𝑖𝜷) (18)

𝑃𝑟(𝐶𝑢𝑙𝑡𝑖𝑣𝑎𝑡𝑒 𝐻𝑉𝐴) = 𝑃𝑟(𝑐1𝑖 = 0|𝑿𝟏𝒊) = 1 − Φ(𝑿1𝑖𝜷) (19)

f(𝑐1𝑖|𝑿𝟏𝒊) = (1 − Φ(𝑿𝟏𝒊 𝜷)1[𝑐1𝑖=0](Φ(𝑿𝟏𝒊 𝜷)

1[𝑐1𝑖=1] (20)


43

Where 𝑦1 represents the household’s production levels if the household produces HVA. The

1 in 𝑐1𝑖 represents the stage one decision for any given household, i, to cultivate HVA given

𝑿𝑖𝑡, the hypothesized explanatory factors that influence this decision, and 𝜷 represents a

vector of the coefficients to be estimated in stage one. Equation (18) represents the

probability that the household does produce, whereas (19) represents the probability that it

does not. Equation (20) demonstrates the Bernoulli distribution showing equations (18) and

(19) in the same equation. Conditional on households passing the first stage hurdle, they

reach the second stage, so that all of the population that proceeds are HVA producers.

Stage 2. Market Participation Decision

As the decision tree indicates, in the second stage, households are separated between net

buyers, autarkic, and net sellers, all of whom are HVA producers. Conditional on the

household being a HVA producer, they reach the second stage decision; the decision to

participate in the market and to what extent. There are three outcomes this study explores in

stage two which are the probability of the following; 1) being a net buyer 2) autarkic (does

not participate) or, 3) be a net seller in the market. Like the stage one decision, a method that

measures the probability of these outcomes occurring, based on the hypothesized factors that

influence this decision is an appropriate method to address the research question. In stage

two, an ordered probit is the elected method that predicts the probability of a household

falling within one of the three specified market participation categories. Much of the same

conceptual framework applies to the ordered probit as in the standard probit, although to

include an additional outcome. Consider the latent variable model;

𝑦∗ = 𝑿𝜷𝟐 + 𝑒 (21)

The observable criteria for the outcome values that 𝑦∗ can take, m are defined as;

𝑦∗ = 𝑚 if 𝛼𝑚−1 ≤ 𝑦∗ ≤ 𝛼𝑚 for 𝑚 = 1, … ,𝑀 (22)

Where 𝛼0 < 𝛼1 < 𝛼2 <. . . . < 𝛼𝑀 and 𝛼0 = −∞ and 𝛼𝑀 = ∞ (23)

The alphas, 𝛼 denote the thresholds that define the cutoff points for the ordinal values of the

dependent variable. Therefore, given the three outcomes, the predicted probabilities of each

ordinal outcome is;

𝑃𝑟(𝑦 = 0) = 𝑝(𝑦∗ ≤ 𝛼0) = Φ(𝛼0- 𝑿𝜷𝟐) (24)

𝑃𝑟(𝑦 = 1) = 𝑝(𝛼0 ≤ 𝑦∗ ≤ 𝛼1) = Φ(𝛼1- 𝑿𝜷𝟐) - Φ(𝛼0- 𝑿𝜷𝟐) (25)


44

𝑃𝑟(𝑦 = 2) = 𝑝(𝛼1 ≤ 𝑦∗) = 1 - Φ(𝛼1- 𝑿𝜷𝟐) (26)

The full distribution of 𝑦𝜃 in the ordered probit can then be seen as;

𝑓(𝑦∗|𝑿) = [Φ(𝛼0 − 𝑿𝜷𝟐)]1[𝑦=0] ∗ [ Φ(𝛼1 − 𝑿𝜷𝟐) − Φ(𝛼0 − 𝑿𝜷𝟐)]

1[𝑦=1]

∗ [ 1 − Φ(𝛼1 − 𝑿𝜷𝟐)]1[𝑦=2]

(27)

Like in the standard probit, the parameters from an ordered probit regression should be

limited to the sign of the parameters. The interpretation from the ordered probit regression

differ from the probit, because they include a third outcome. Given that net buyers are equal

to zero, autarkic are equal to one, and net sellers are equal to two, increasing values of the

variables with positive coefficients indicate that the observations are more likely to be

autarkic than net buyers, and more likely to be a net seller than autarkic (Burke et al. 2015).

The marginal effects of the individual parameters can also be calculated for the ordered probit

as in the standard probit (Equation 15), however, they are derived with respect to a specific

outcome.

Applying this to triple hurdle framework, the notation proceeds with 𝑝2𝑖 which will represent

the stage two outcomes and 𝑦2 indicate the quantity of market purchases. The following will

denote the three outcomes of market participation;

𝑝2𝑖 = 0[𝑦1 − 𝑦2 < 0] (28)

𝑝2𝑖 = 1[𝑦1 − 𝑦2 = 0] (29)

𝑝2𝑖 = 2[𝑦1 − 𝑦2 > 0] (30)

Equation (21) represents net buyers because the quantity purchased in the market for HVA

producers, 𝑦2 is greater than the quantity sold for HVA producers, 𝑦1, therefore 𝑝2𝑖 takes on

the value of zero; equation (29) indicates the household is autarkic and does not participate

and 𝑝2𝑖 is equal to one; while equation (30) shows net sellers who sell more than they

purchase and 𝑝2𝑖 takes on the value of two. As mentioned by Burke et al. (2015), for

estimation purposes for the ordered probit, it is necessary to have net buyers and net sellers

on opposite sides of autarky. In stage two, we can define the latent variable as;

𝑝2𝑖∗ = 𝑿𝟐𝜷𝟐 + 𝑒 𝑒|𝑥2~𝑛𝑜𝑟𝑚𝑎𝑙 (31)


45

Where 𝑿𝟐 represents a vector of the explanatory variables determining the stage two outcome

for market participation. According to Burke et al. (2015) the specifications in equations

(28)-(30) will define the outcome of the latent variable within the specified thresholds

denoted as 𝛼 such that;

𝑝2𝑖 = 0 𝑖𝑓 𝑝2𝑖∗ < 𝛼0

(32)

𝑝2𝑖 = 1 𝑖𝑓 𝛼0 < 𝑝2𝑖∗ < 𝛼1

(33)

𝑝2𝑖 = 2 𝑖𝑓 𝛼1 < 𝑝2𝑖∗

(34)

Now that the outcomes are specified, notation can be derived that shows the probability of a

particular market participation outcome;

𝑝𝑟(𝑝2𝑖 = 0|𝑿𝟐, 𝛼, 𝜷𝟐 ) = 𝑝𝑟(𝑝2𝑖∗ < 𝛼0|𝑿𝟐) = Φ(𝛼0 − 𝑿𝟐𝜷𝟐) (35)

𝑝𝑟(𝑝2𝑖 = 1|𝑿𝟐, 𝛼, 𝜷𝟐 ) = 𝑝𝑟(𝛼0 < 𝑝2𝑖∗ < 𝛼1|𝑿𝟐) = Φ(𝛼1 − 𝑿𝟐𝜷𝟐)1 −Φ(𝛼0 − 𝑿𝟐𝜷𝟐) (36)

𝑝𝑟(𝑝2𝑖 = 2|𝑿𝟐, 𝛼, 𝜷𝟐 ) = 𝑝𝑟(𝛼1 < 𝑝2𝑖∗ |𝑿𝟐) = 1 − Φ(𝛼1 − 𝑿𝟐𝜷𝟐) (37)

The full distribution for the stage two ordered probit in triple-hurdle notation can then be seen

as;

𝑓(𝑝2𝑖|𝑿𝟐) = [Φ(𝛼0 − 𝑿𝟐𝜷𝟐)]1[𝑝1𝑖=0] [Φ(𝛼1 − 𝑿𝟐𝜷𝟐)1 − Φ(𝛼0 −

𝑿𝟐𝜷𝟐)]1[𝑝1𝑖=1][1 − Φ(𝛼1 − 𝑿𝟐𝜷𝟐)]

1[𝑝1𝑖=2]

(38)

Stage 3. Degree of Market Participation

Thus far, households have faced the first hurdle which separated households into HVA

producers and non-producers. Those that produce proceed to the second stage which

determines their market participation outcome. Autarkic households (non-market

participants), do not proceed to the third hurdle, which measures either net sales if the

household is a net seller, or net purchases if the household is a net buyer. Unlike the first two

stages, the estimation method will require a model that measures continuous outcomes.

Additionally, the method to address continuous outcomes must deal with the truncated nature

of the data where whole observations are missing depending on how they have been excluded

from the hurdle model. Therefore, the third stage outcome uses log-normal regressions to

measure net sales or net purchases.

As seen in equations (21)-(23), the net buyers and net sellers are defined by total market sales

less total market purchases of HVA producers, where 𝑦3 represents net buyers quantity


46

bought as 𝑦3 = 𝑦2 − 𝑦1, if 𝑦1 − 𝑦2 < 0, and net sellers, 𝑦4 seen as 𝑦4 = 𝑦1 − 𝑦2, if 𝑦1 −

𝑦2 > 0. Like the previous two stages, the log-normal equations derive their parameters using

maximum-likelihood estimation. First, the distribution of 𝑦3 and 𝑦4 given 𝑿𝟑 can be derived

as;

𝑓(𝑦3|𝑿𝟑, 𝛿𝟑) = ∅ [log(𝑦3) − 𝑿𝟑, 𝛿𝟑

𝜎𝟑] /(𝑦3𝜎

𝟑) (39)

𝑓(𝑦4|𝑿𝟒, 𝛿𝟒) = ∅ [log(𝑦4) − 𝑿𝟒, 𝛿𝟒

𝜎𝟒] /(𝑦4𝜎

𝟒) (40)

Where 𝑦3 and 𝑦4 are the net quantities bought and sold and 𝑿𝟑 are the stage three

explanatory variables and its parameters to be estimated, ∅ is the standard normal probability

density function and 𝜎𝑗 is the standard deviation of the error term for the random process

determining 𝑦𝑗 from each of the equations. The log transformation occurs within the

likelihood function, so 𝑦3 and 𝑦4 are the levels of net quantities. Since the dependent variable

in stage three is in log-form, the interpretation of the coefficients also in log-form should be,

with a one percent increase in the explanatory variable we should expect a percentage change

in the dependent variable equivalent to the coefficient. For explanatory variables not in log

form, a one unit change in the explanatory variable changes the dependent variable by the

coefficient multiplied by 100, expressed as a percentage. A one unit change in a given

variable is associated with a predicted change in the dependent variable by 𝛽 ∗ 100. As in

Burke et al. (2015), the explanatory variables in the estimations of each stage remain the

same for the structural derivation of the triple-hurdle model. The next section outlines the

econometric estimations of the conceptual framework.

4.2 Econometric Estimation

Under standard assumptions maximizing the likelihood function produces efficient, unbiased

and consistent estimates of the triple-hurdle model parameters. Maximum likelihood

estimates the values of a set of parameters of a model from a sample population using a

function that creates a likelihood distribution. Maximum likelihood estimates are the

parameter values that are most likely to have produced the observed data. Observing an

independent sample 𝑥1, 𝑥2, … , 𝑥𝑛 the log-likelihood function can be applied to various

distributions, whether it be for discrete outcomes or continuous. If the observations are


47

independent and identically distributed, the log likelihood function for a continuous random

variable can be seen as;

𝑙𝑜𝑔𝐿(𝜃) = 𝑙(𝜃) = −𝑛

2log(2𝜋𝜃2) −

1

2𝜃2∑(𝑥𝑖 −

𝑛

𝑖=1

𝜃1)2

(41)

(Wooldridge 2013). Estimating the log-likelihood parameters involves maximizing this

equation by setting it equal to zero and taking the first order conditions of 𝑥𝑖 with respect to

𝜃. This demonstrates the derivation of a log-likelihood function for a continuous random

variable which will be the case in the third hurdle for net sales and net purchases volumes.

However, MLE can be applied to a distribution of discrete variables, as well, which will be

demonstrated in the estimation strategy for the parameters in the probit and ordered probit

from the first two stages.

Given the nature of this research question, the likelihood function integrates all three stage’s

models into one. The rationale for creating one likelihood function for all stages is to

understand how the effects from one stage can affect subsequent stage outcomes. For

example, how do the effects of increased rainfall change the probability of being a net selling

HVA producer, but also effect the expected value of sales for a given household?

HVA production and market participation decisions are hypothesized to be a function of

demographic, farm and regional characteristics. These factors will remain the same across all

stages of the triple hurdle estimations. For notational purposes, 𝑿𝟏, 𝑿𝟐, 𝑿3 and 𝑿4 will

denote a vector of the demographic, farm and community characteristics in the corresponding

stage denoted by its subscript. 𝑿𝟏 denotes the factors determining stage one production

decision, 𝑿𝟐 represents the factors determining market participation decision, while

𝑿𝟑 and 𝑿𝟒 are factors determining net quantities bought for net purchasers and net quantities

sold for net sellers. The empirical framework of each separate estimation defined in the

previous section can now be shown within one likelihood function. The following notation

𝑓(𝒄, 𝒑, 𝒚|𝛾1, 𝛾2, 𝛽, 𝜌, 𝛿3, 𝛿4, 𝜎3, 𝜎4) represents the joint distribution function of the three

stages outcomes given the parameters of each stage to be estimated. Thus, as in Burke et al.

(2015), the full likelihood distribution and the log-likelihood function used to estimate the

parameters, 𝜃 can be specified for any observation, i as;

Maximum Likelihood Function:


48

𝑓(𝒄, 𝒑, 𝒚|𝛼1, 𝛼2, 𝛽, 𝜌, 𝛿3, 𝛿4, 𝜎3, 𝜎4) = (42)

[1 − Φ(𝑿1𝜷]1[𝑖𝑓 𝑐1=0] ∗ [Φ(𝑿1𝜷)

[Φ(𝛼0 − 𝑿2𝜷)∅(

[ln(𝑦3) − 𝑿3𝜹3]𝜎3

)

𝑦3𝜎3]

1[𝑖𝑓 𝑝2=0]

[Φ(𝛼1 − 𝑿2𝜷) − 𝜑(𝛼0 − 𝑿2𝜷)]1[𝑖𝑓 𝑝2=1]

[Φ(𝑿2𝛽 − 𝛼1)∅ (

[ln(𝑦4) − 𝑿4𝜹4]𝜎4

)

𝑦4𝜎4]

1[𝑖𝑓 𝑝2=2]

]1[𝑖𝑓 𝑐1=1]

Log-Likelihood Function:

𝑙𝑖(𝜃) =

1[𝑐1𝑖 = 0] log[1 −Φ(𝑿1𝑖𝜷)] + 1[𝑐1𝑖 = 1] log[Φ(𝑿1𝑖𝜷)]

+ 1[𝑝2𝑖 = 0] (log [Φ(𝛼0 − 𝑿2𝑖𝜷)] − log (𝑦3𝑖) − (1

2)log (𝜎3

2) − (1

2)log (2𝜋)

−1

2([log(𝑦3𝑖) − 𝑿3𝑖𝜹3

2

𝜎32 ])

+ 1[𝑝1𝑖 = 1] log[Φ(𝛼1 − 𝑿2𝑖𝜷) − Φ(𝛼0 − 𝑿2𝑖𝜷)]

+ 1[𝑝1𝑖 = 2] log[ 1 − Φ(𝛼1 − 𝑿2𝑖𝜷) − log(𝑦4𝑖) − (1

2) log(𝜎4

2) − (1

2) log(2𝜋)

−1

2([log(𝑦4𝑖) − 𝑿4𝑖𝜹4

2

𝜎42 ])

(43)

Here, Φ(∙) represents the standard normal cumulative density function; ∅ denotes the

standard normal probability density function; 𝜌 represents the stage one parameters of 𝑿1;

𝛼0 and 𝛼1 are used to represent the limits in order to define the categorical outcomes of the

ordered probit; 𝛽 represents the parameters for stage two on 𝑿2; 𝛿3 and 𝛿4 represent the

parameters for the third stage for 𝑦3 and 𝑦4. Where it reads, 𝑖𝑓 𝑝2 = 0, 1, or 2, this refers to

the ordinal outcome of zero to represent net buyers, one for autarkic, and two denotes net

sellers. Lastly, 𝜎3 and 𝜎4 are the error variance for the third stage net quantities bought and

sold, ln(𝑦3) and ln(𝑦4).

Burke et al. (2015) points out that the log-likelihood equation (43) can be estimated

simultaneously, as well as separately due to the separability feature of the likelihood function.


49

This allows for each stage to be estimated separately. If estimated separately, the following

steps are required;

Estimation Steps:

1. 𝜌 can be estimated using a probit regression of 𝑐1 on 𝑿1;

2. 𝛽 can be estimated using an ordered probit by regressing 𝑝2 on 𝑿𝟐 if 𝑐1 = 1;

3. 𝛿3 can be estimated by regressing log(𝑦)3 on 𝑿𝟑 using only the observations from

𝑝2 = 0;

4. 𝛿4 can be calculated by regressing log(𝑦)4 on 𝑿𝟒 using the observations from 𝑝2 = 2.

For valid inference of standard errors using separate ML estimation, Burke stipulates that the

error terms in all equations are uncorrelated conditional on all explanatory variables. If the

condition does not hold, separately estimating the equations will result in biased coefficients.

The test for this condition follows a Heckman test for selection bias (Burke et al. 2015;

Wooldridge 2010). The test proceeds by predicting the Inverse Mills Ratio (IMR) around the

dependent variable outcome and including it as an explanatory variable in the estimate of the

subsequent stage model. For example, the 𝐼𝑀𝑅𝑐 will be calculated around the probability of

being a producer of HVA in stage one, the 𝐼𝑀𝑅𝑐 will then be included in the second stage

ordered probit as an explanatory variable. If the errors are conditionally uncorrelated, 𝐼𝑀𝑅𝑐

should have no explanatory power in the second stage estimation. A simple t-test can test

this condition under the hypothesis that 𝐼𝑀𝑅𝑐 = 0. If its coefficient is statistically different

than zero, the null hypothesis is rejected and estimates would be biased if the IMR were

excluded. If the null cannot be rejected, the other estimates are only unbiased and consistent

when the IMR is included, but standard statistical inference is not valid because you have

included a predicted regressor. Nevertheless, when the null cannot be rejected, then one can

proceed to the second stage. The same process is implemented on the next stage, gathering

the IMR from the stage two ordered probit around the probability of being a net seller or net

buyer and estimating them as an explanatory variable in the following stage equation using

ordinary least squares. The same t-test procedure is applied at stage two to determine if

inference can be made without bias. It should be noted that these procedures are conducted

using only the appropriate observations that carry over to subsequent stages.

Given the data structure used in the study, simultaneous maximum likelihood estimation of

all three stages does not converge for parameter estimates. Therefore, each stage is estimated


50

separately and satisfies the assumption of uncorrelated errors between all stages for valid

inference. Burke notes that interpretation of each explanatory variable depends on the effects

from all equations throughout the model. For example, for any observation, four

unconditional probabilities can be calculated, as can the partial effect on these probabilities

with respect to each explanatory variable. Among the many outcomes of interest in the

triple-hurdle model, key results include:

𝑃𝑟(𝑐𝑢𝑙𝑡𝑖𝑣𝑎𝑡𝑖𝑛𝑔 𝐻𝑉𝐴) = 𝑃𝑟(𝑐1𝑖 = 0|𝑿𝒊) = 1 − Φ(𝑿1𝑖𝝆) (44)

𝑃𝑟(𝑛𝑒𝑡 𝑏𝑢𝑦𝑖𝑛𝑔 𝑝𝑟𝑜𝑑𝑢𝑐𝑒𝑟) = Pr(𝑐1𝑖 = 1, 𝑝2𝑖 = 0|𝑿𝒊) = Φ(𝑿1𝑖𝝆)Φ(𝛼0 −𝑿2𝑖𝜷) (45)

𝑃𝑟(𝑎𝑢𝑡𝑎𝑟𝑘𝑖𝑐 𝑝𝑟𝑜𝑑𝑢𝑐𝑒𝑟) = Pr(𝑐1𝑖 = 1, 𝑝2𝑖 = 1|𝑿𝒊) = Φ(𝑿1𝑖𝝆)(Φ(𝛼1 − 𝑿2𝑖𝜷)Φ(𝛼0 − 𝑿2𝑖𝜷) (46)

Pr(𝑛𝑒𝑡 𝑠𝑒𝑙𝑙𝑖𝑛𝑔 𝑝𝑟𝑜𝑑𝑢𝑐𝑒𝑟) = Pr(𝑐1𝑖 = 1, 𝑝2𝑖 = 2|𝑿𝒊) = Φ(𝑿1𝑖𝝆)(1 − Φ(𝛼1 −𝑿2𝑖𝜷)) (47)

In addition to the abovementioned probabilities, unconditional expected values can be

calculated from the likelihood function for net sales and net purchases for all observations.

Burke explains how this implies the expected values are not conditional on the values of the

dependent variables, such as the household being a producer or a market participant (Burke et

al. 2015). For each observation, the derivation can be seen as:

𝐸(𝑛𝑒𝑡 𝑝𝑢𝑟𝑐ℎ𝑎𝑠𝑒𝑠) = 𝐸(𝑦3𝑖|𝑿𝑖) = Φ(𝑿1𝑖𝝆)Φ(𝛼0 − 𝑿2𝑖𝜷) ∗ 𝑒𝑥𝑝(𝑿3𝑖𝛿3 +

𝜎32/2)

(48)

𝐸(𝑛𝑒𝑡 𝑠𝑎𝑙𝑒𝑠) = 𝐸(𝑦4𝑖|𝑿𝑖) = Φ(𝑿1𝑖𝝆)Φ(𝑿2𝑖𝜷 − 𝛼1) ∗ 𝑒𝑥𝑝(𝑿4𝑖𝛿4 + 𝜎42/2) (49)

Average partial effects are calculated for the first two stages, while the third stage log-normal

regression output can be interpreted as conditional average partial effects, according to Burke

et al. (2015).

The last quantities of interest are the partial effects for each stage of the model. Probit

models traditionally base inference on marginal effects because of the non-linear nature of

the model, although partial effects are calculated because they are derived considering

hypothetical, interaction terms. The partial effect depends on parameters and explanatory

variables from all of the stages. These can be derived as follows:


51

𝜕𝐸(𝑦4𝑖|𝑿𝒊)𝜕𝑥𝑘𝑖

= 𝝆𝑘∅(𝑿1𝑖𝝆)Φ(𝑿2𝑖𝜷 − 𝛼1) ∗ exp (𝑿4𝑖𝛿4 +𝜎42

2) + 𝛽𝑘Φ(𝑿𝟏𝒊𝝆)∅(𝑿2𝑖𝜷−

𝛼1) ∗ exp(𝑿4𝑖𝛿4 + 𝜎42/2) + 𝛿4𝑘Φ(𝑿1𝑖𝝆)Φ(𝑋2𝑖𝜷 − 𝛼1) ∗ exp (𝑿4𝑖𝛿4 +

𝜎42

2)

(50)

From equation 50, we can derive the partial effect of a continuous variable, denoted at 𝑿𝑘, on

the unconditional expected value of net sales or net purchases. For ease of interpretation, the

partial effects are averaged across the sample, including the partial effects on the

unconditional expected value, as in Burke et al.(2015). Additionally, we can estimate the

partial effects of each explanatory variable on the probability of an HVA producing

household being a net seller. The standard errors of the average partial effects are calculated

using the bootstrap technique that relies on random sampling with replacement.

Estimation of Burke’s et al. (2015) triple-hurdle model applied to panel data uses a pooled

estimator with maximum likelihood estimation. When performing the estimations,

households are clustered to provide robust standard errors, which will account for

autocorrelation resulting from inter-temporal dependence between observations.

Hypotheses:

This research hypothesizes that,

1) Access to information, infrastructure, inputs, capital, and crop prices, are

significant determinants of HVA cultivation and market participation,

2) The same processes that influence a farmer to cultivate HVA increase the

probability of being a net seller in the market.

The hypotheses will be tested based on the statistical significance of the coefficients from the

model estimations being different than zero. The conclusions to be drawn from the results

that test these hypotheses will provide a deeper understanding of whether common policy

approaches to promote market participation may have further-reaching effects than

anticipated. Dually, this can inform policy-makers if policies should be implemented that

focus on market participation and the adoption of HVA separately.

Based on previous literature, access to credit, inputs, information, capital and infrastructure

constraints are the key hypothesized determinants of crop diversification. Since Uganda has

tried to promote policies that broaden its export base and increase smallholder

commercialization, exposure to extension services which is believed to transfer such


52

information, is hypothesized to be a key determinant of diversification. Many HVA products

are considered capital intensive, therefore access credit may facilitate the startup costs of

commencing HVA production. Ownership of productive technologies has also been

considered an enabling factor in the decision to commence for the same reason. Households

that do not use improved seed have increased production risk and may choose to prioritize

home consumption through few crops, therefore its adoption is believed to have a positive

relationship to crop diversity. Hired labor, another input, which increases household working

capacity, is also widely cited to positively impact diversity. Table 4.1 shows expected signs

on HVA production decision and its probability of being a net seller, conditional on being a

HVA producer:

Table 4.1: Expected Sign of Coefficients on Dependent Variable

Explanatory Variable Decision to Cultivate HVA Pr(Net Seller)

Ln(land) + +

Male head + +

Head education + +

Head age - +

Received credit + +

Ln(productive assets val.) + +

Urban area - -

Distance to market - -

Distance to road + -

Distance to population center - -

Tech endowment + +

Improved seed + +

Annual average rainfall + +

Plot elevation + -

Hired labor + +

Adult equivalents + +

Ln(Dist. Beans price) - +

Ln(Dist. Cassava price) - +

Ln(Dist. Maize price) - +

Ln(Dist. Coffee price) + +


53

CHAPTER 5: DATA AND VARIABLES DESCRIPTION

5.1 Dependent Variables

The triple-hurdle model, will use a total of four different dependent variables. The first being

the decision to cultivate HVA, representing a binary decision of zero to indicate a non-

producer of HVA and one, to indicate that the household does produce HVA. Stage two

outcome uses an ordered probit, taking on three values; zero, one, or two to sort households

into net buyers, autarkic, and net-sellers. Third stage dependent variables are net quantities

of purchases and sales, both in log-form. Table 5.1 shows their definitions and the values

which they can take on, as well as their mean sample values at the starting panel wave in

2009 and last wave in 2012.

Table 5.1: Dependent Variables

Dependent Variables Definition Mean 2009 Mean 2012

Probit: HVA Pr(𝐶𝑢𝑙𝑡𝑖𝑣𝑎𝑡𝑒 𝐻𝑉𝑃 = 0, 1) 0.524 0.609

Ordered Probit: NS, A, NB Pr(𝑁𝑆, 𝐴𝑢𝑡, 𝑁𝐵) = 0,1,2 1.649 1.801

Ln(Net Purchases) ln(𝑠𝑎𝑙𝑒𝑠 − 𝑝𝑢𝑟𝑐ℎ𝑎𝑠𝑒𝑠) ∗ −1 11.761 12.236

Ln(Net Seller) ln(𝑠𝑎𝑙𝑒𝑠 − 𝑝𝑢𝑟𝑐ℎ𝑎𝑠𝑒𝑠) 12.163 12.522

Source: World Bank LSMS n=1,968 in 2009/2010, n = 2,117 and n = 2,032 in 2011/2012

Table 5.1 shows that the probit takes on the values of either 0 or 1, to indicate if the

household produces HVA. The ordered probit can take the values of 0, 1, 2 to separate

households into their respective market participation categories. Ln(net purchases) shows the

log of sales minus purchases multiplied by -1 to obtain a positive value if the household is a

net purchaser in the market and HVA producer. Ln(net sales) provides the log of sales minus

purchases if the household is a net seller and an HVA producer. The mean values of the

probit from 2009 to 2012 show that 52.4% of the sample were HVA producers, while in 2012

60.9% became producers. The ordered probit mean values show that a there was an increase

in the number of net sellers from 2009 to 2012 illustrated by the increase from 1.64 to 1.80.

Of net sellers and net buyers who produce HVA, net sellers sold more and net buyers bought

more from 2009 to 2012.

The stage one probit that separates HVA producers and non-producers depends on a specified

set of HVA products. HVA are considered cash crops, as well as horticulture. One crop that


54

is an exception in the horticulture list are plantains, which are a staple crop of Uganda.

Plantains are cultivated by 47% of all farmers and provide the greatest amount of caloric

intake of all Ugandan crops. Their prevalence makes them a crop with low economic returns

in Uganda and are less frequently traded. Given the nature of the study – to analyze the

decision to cultivate HVA and its marketing outcomes, they are left out of the selected list

HVA products.

From the three waves of survey panel data from the World Bank’s LSMS 2009/2010,

2010/2011, and 2011/2012, Ugandan farmers who cultivate any of the following crops are

considered HVA producers: cabbage, tomatoes, carrots, onions, pumpkins, dodo, eggplants,

sugar cane, cotton, tobacco, oranges, paw paw, pineapples, mango, jackfruit, avocado,

passionfruit, coffee, cocoa, tea, ginger, curry, oil palm, vanilla, black wattle. As indicated in

the table, in 2009 35.1% are HVA producers, while 37.8% are producers in 2012. Contrarily,

crops that do not constitute HVA from the survey data are maize, finger millet, sorghum,

wheat, barley, rice, field peas, cow peas, chick peas, pigeon peas, soyabeans, sunflower,

sesame, yam, coco yam, and Irish potatoes, beans, banana, cassava, sweet potatoes, and

groundnuts. Over the panel years, market participation of sellers has also increased. Over

the time period of the sample, 2009-2012, households marketing their products has shown a

gradual increase from 74.4% to 80.2%.

5.2. Explanatory variables

Conventional wisdom argues that agricultural household decisions are a function of

household (demographic), farm, and community characteristics (Mofya-Mukuka et al. 2016).

Other studies such as Turner (2014) and Burke et al. (2015) have divided the determinants

into different sub-categories, such as household and farm-level characteristics, market prices

and market constraints. For an in depth view of the explanatory variables used in these

analyses, determinants are divided into household, farm, and regional characteristics, along

with market prices and market constraints. All of the used variables and their calculations are

from the data offered from the World Bank’s LSMS survey of Uganda 2009-2012. The

explanatory variables used in the analyses are:

Household-Level Characteristics:

These variables include those related to the characteristics of the household members such as

human capital and life-cycle stage of the farm-household (Benin 2004). Life cycle of the


55

farm is often associated with the age of household head. This serves as a proxy for farming

experience and managerial skill, but also has implications on the decision to adopt new crops.

It is expected that the age of the household head has a direct relationship with farm diversity

because younger households may be more amenable to experiment with different crops and

varieties, where older households may be expected to be the opposite, with greater propensity

to rely on traditional practices (Van Dusen 2000).

The effect of being male versus being female is expected to be positive on crop

diversification. Female farmers generally face greater social barriers than men (Farnworth,

Akamandisa, and Hichaambwa 2011). This is the result of societal gender roles and because

male household heads often obtain higher levels of education which is related to greater

access to farming technology services, credit and selling of produce (FAO) which has

contrasting outcomes for agricultural productivity, as well. The household head’s gender is

captured with use of a dummy variable indicating one if male and zero if female. Similarly,

education level of the household head is positively being expected to be positively associated

with farmer’s ability to access information and to meet and adopt more complex management

requirements for particular crops (Rehima et al. 2013). Another widely used meter of human

capital is the number of full-time adult equivalents in the household which represent available

non-hired labor.

To summarize, the household/demographic characteristics used as explanatory variables in

the estimations are age, gender and level of education of the household head, as well as full-

time adult equivalents.

Farm-Level Characteristics:

Different measures that serve as proxies for household wealth are often cited as factors which

influence crop diversification. This is argued to be the case due to farmers typically needing

some form of support with working capital in order to diversify (Carter 2000) (Mofya-

Mukuka & Hichaambwa 2016). Household income is hypothesized to be associated with

crop diversification decisions, however it presents an endogeneity problem because it is a

function of land allocation. Instead, many agricultural studies use the household’s value of

productive assets in log form. This study uses this approach – log value of owned productive

assets. This is calculated by aggregating all productive assets that could be related to crop

diversification decisions. Assets such as televisions, computers, radios, electronic equipment

may sound irrelevant, but are included in the calculation of productive assets because many


56

farmers obtain information about crops from such sources. Additionally, it includes working

assets such as tractors, motor vehicles, generators, among others are included in the

calculation.

Total household landholdings in log-form is included as an explanatory variable.

Landholdings are expected to positively affect the probability of a household diversifying and

adopting HVA because households with greater amounts of land have more to work with and

are more likely to produce a crop for the market once their consumption needs are met

(Turner 2014). However, a non-linear or, inverse relationship may be present between the

decision to diversify and land size. Larger farms may be expected to specialize because they

have more working capital.

Bellemare (2012) offers several explanations of this inverse relationship. One seminal

account draws from Sen (1966), describing that in developing countries where working

capital is scarce, households with smaller farm sizes simply had more labor per land unit than

larger farms. Another explanation for the inverse relationship is related to the analyst’s

omitted variables problem. Unobserved factors such as soil quality are, on a very rare basis,

included in researcher’s estimations for this relationship. Farmers tend to cultivate their most

fertile land first, then expand cultivation into areas of lesser soil quality which makes larger

farms appear less productive (Bellemare 2012). The last explanation deals with the problem

of a spurious inverse relationship due to measurement error.

Inputs such as improved seed and use of hired labor are used as explanatory variables. Use

of hired labor is an additional variable to represent labor availability, which increases the

working capacity of the household. Since HVA can be capital intensive, additional labor is

hypothesized to facilitate HVA adoption. Households that do not use improved seed can

have increased production risk and may choose to prioritize home consumption through few

crops, therefore its adoption is believed to have a positive relationship to crop diversification

decisions. Both are controlled for with the use of a dummy variable to indicate whether the

household hired labor or used improved seed.

Regional Characteristics:

Annual total average rainfall may have various outcomes on crop diversification decisions.

Literature suggests that farmers in developing countries are often risk averse and crop

diversification is a commonly cited method to hedge against price and production risk

(Bezabih and Sarr 2012). This decision comes from insuring that the household can have


57

income from market-sold crops in the event that household production is low. Annual total

average rainfall at the district level is used for the analysis.

Population density in developing countries can have varying effects for agricultural

productivity and diversification. Uganda is the ninth most densely populated country in

Africa, therefore urban areas, especially are expected to have high density. Consequently,

households in urban versus rural areas can have different outcomes with respect to crop

choices. Additionally, in urban areas, it is more likely that households have members

engaged in off-farm labor. Birthal et al. (2007) states that a household’s occupation is a key

determinant of crop choices. Households with off-farm occupations may be more likely to

cultivate crops that are less labor-intensive due to competing uses of labor. Like variable

rainfall, the decision to venture into HVA may play a role as an insurance mechanism against

variable production in highly densely populated areas and less of a role in areas of less

density. A binary indicator of rural and urban households is included, but does not fully

address true population density.

Uganda is mountainous and elevations can vary, which makes some crops more feasible than

others (Gerold 2004). In areas that have agroecological features less suitable for crop

production, such as regions of variable rainfall or in high elevations, crop diversification is

seen as a risk mitigation strategy by diversifying into crops that use less resources.

Therefore, greater elevation (meters above sea level) are believed to be positively related to

crop diversification decisions. Arabica coffee, for instance, is grown at higher elevations of

Uganda and may therefore positively influence the probability of farmers being an HVA

producer. Elevation is reported at the household-level.

Market Prices:

Changing diversification patterns is widely believed to reflect changing infrastructure and

market signals (Kurosaki 2003). Higher profitability is the key driver to induce crop

diversification. Perfect knowledge is not available to farmers and expected prices are a

widely used proxy used for price, although this analysis uses current prices in order to retain

all panel waves. Average maize, cassava, coffee, and beans prices at the district-level are

included. This may represent an endogeneity problem by including a portion of the

household’s price in the district calculation because individual behavior can influence its own

selling price, but it is likely to be largely reduced by using the district average (Turner 2014).

Maize, cassava, and beans, which are substitutes for HVA, are produced with high frequency


58

in the sample and their prices are hypothesized to have inverse relationships with the

probability of cultivation HVA and the HHI. These variables are calculated by averaging the

amount received per kilogram at the district-level and are in log-form to improve the data

fit.to reduce skewness in the distribution to improve data fit.

Constraints:

Returns to crop diversification depend on the accessibility and availability of various

infrastructure facilities. Irrigation, electricity, transportation, storage and accessibility of

markets encompass the bulk of infrastructure constraints with respect to diversification (Abro

2012). This analysis uses distance to the nearest market, urban center and road as the

constraining infrastructure characteristics. These partially represent transfer costs, which

include transaction costs plus transportation costs. For example, if a household is close to a

market and has a paved road to access it, its transport costs are greatly reduced compared to

the isolated farmer without poor travel passages. Transaction costs, which are associated

with institutional characteristics of the market are not captured in this data. Incentives to

become more efficient and increase agricultural productivity become greater when transfer

costs are reduced, believed to incentivize HVA production.

Birthal (2007) highlights the importance of extension services because, unlike staple crops,

information requirements of HVA products tend to be high. Obtaining access to information

about different crops and management practices encourage diversification and supplies

farmers with new farming technologies that impact productivity. For analysis, only extension

services, which are often offered by agricultural extension networks is used as an explanatory

variable, but radios, television and other information channels have been used before.

Extension service contact may be impacted by the infrastructure constraints already in place,

presenting an endogeneity problem. However, it is unlikely that there are significant changes

in the infrastructure development across the panel waves, which should not cause bias in the

estimates, although both are included to avoid any omitted variable problems.

Financial services are often scarce for many rural farmers in lesser developed countries

(Birthal et al. 2007). Often times, such is needed to purchase or make in initial investment in

inputs in order to change production. Moreover, the cultivation of many HVA crops is

capital intensive and obtaining credit may facilitate the high start-up costs. Credit access

should thus be positively related with the uptake of HVA. The share of households in a

district receiving credit is used as a proxy for credit availability (Burke et al. 2015). If a


59

household receives any form of credit from an official agency, they qualify as a credit

receiving household. Credit-receiving households are aggregated at the district-level and

divided by the aggregate number of households in the district.

5.3. Descriptive Statistics

Below in table 5.2 provides a general overview of the main variables that will be used in the

econometric modelling. Their average values across, standard deviations, minimum and

maximum values are reported across the entire sample. UGX denotes Ugandan Shillings,

MASL denotes meters above sea level, Ha is hectares and Km represents kilometers.

Table 5.2: Dependent Variables, Household and Regional Characteristics

Variable Unit Obs. Average Values Std. Dev. Min Max

Dependent Variables

HVP Producer (1 = yes) 6,104 0.36 0.5 0 1

NS, Aut. NB Category 8,688 1.8 0.5 0 2

Ln(Net Sales) UGX 4,110 12.4 1.6 3 18.4

Ln(Net Purchases) UGX 442 12.1 1.3 7.9 15.4

Household and Farm Characteristics

Male Head (1=yes) 7,113 0.7 0.5 0 2

Ln(Household Land) Ha 5,593 0.8 0.6 0 5.8

Adult Equiv. Adults 7,113 3.2 1.9 0 20

Ln(Prod. Assets) UGX 6,190 9.9 1.6 0 15.2

Hired Labor (1 = yes) 6,603 0.6 0.5 0 1

Improved Seed (1 = yes) 5,915 0.1 0.3 0 1

Tech Endowment (1 = yes) 5,657 0.4 0.5 0 1

Head Age Years 7,111 46.8 15.1 14 100

Head Education Years 5,703 23.7 15.8 10 99

Regional Characteristics

Received Credit (%) 8,898 0.1 0.1 0 0.7

Urban (1 = yes) 7,129 0.2 0.4 0 1

Distance Pop Center Km 7,107 22.5 18.2 0.1 102

Distance to Market Km 7,107 30.6 19.8 0.1 116.2

Distance to Road Km 7,107 7.5 7.5 0 41.4

Ann Avg. Rainfall mm/year 7,107 1129.7 173.7 514 1655

Elevation (MASL) 7,107 1234.5 238.7 621 2396

Ln(Dist. Coff. price) UGX 3,623 8.4 2.9 2.8 14.4

Ln(Dist. Beans price) UGX 4,999 8 3.1 2.5 13.6

Ln(Dist. Maize price) UGX 5,976 7.8 3.3 1.9 14.2

Ln(Dist. Cass. price) UGX 4,723 5.4 2 0.8 13.8

Source: World Bank LSMS 2009-2012. n varies based on data availability. (Head education is on a scale

unequal to grade level). (UGX stands for Uganda Shilling)


60

5.4 Data

This research uses nationally representative data for Uganda from the World Bank’s Living

Standards Measurement Study – Integrated Surveys on Agriculture (LSMS-ISA), which was

implemented in conjunction with The Uganda National Panel Survey (UNPS) and Uganda

Bureau of Statistics (UBOS). Data consists of three panel waves from 2009 to 2012 where

each panel wave contains data over a twelve-month period. Each panel wave is broken into

2009/2010, 2010/2011, and 2011/2012.

The LSMS surveys cover a variety of topics on demographics, labor, income and

consumption dynamics at the household-level, among other welfare indicators of Uganda. Its

survey aims have been primarily to provide annual information to monitor various economic

indicators used to inform and evaluate policy decisions. Due to attrition, new households are

added annually which results in an unbalanced panel set. This study uses the household

survey, agricultural survey and the geospatial data provided. Much of Uganda has two rainy

periods and calculations were aggregated on an annual basis.

In 2005/2006, a baseline survey conducted by the Uganda National Household Survey

(UNHS) with the UNPS interviewed 3,123 households distributed across 322 enumeration

areas (EA) selected from a larger pool of 783 enumeration areas. In the 2009/2010 survey,

the sample size became 2,975, in the 2010/2011 wave is 2,716, and in 2011/2012 is 2,850.

Within each stratum, the UNPS EAs were randomly selected from UNHS EAs with equal

probability to comprise a nationally representative sample. The strata of representativeness

from the UNPS includes Kampala City, other urban areas, central rural, eastern rural, western

rural, and northern rural (World Bank LSMS).

Attrition:

Survey attrition, if not adequately accounted for, can result in biased estimates because they

are not representative of the national sample. Many longitudinal household surveys are

troubled with attrition, especially in developing countries where communication

infrastructure is less advanced, making it difficult to precisely track all the household

respondents across the panel waves, as is the case in the dataset. In the data set, it was

reported from the LSMS that attrition arose primarily from households moving and being

unable to locate, mortality, disintegration (household members part ways and none of them

remain at the original location), or refusal of interview.


61

For tracking of the households in all years, the aim was to identify and survey all 3,123

original panel households. In the baseline survey, conducted in 2005/2006, two households

in each EA were randomly selected to track baseline individuals that moved away from their

original locations since 2005/2006. The UNPS tracked the sub-sample with the intention to

recalibrate the size and composition of the original sample to compensate for attrition. The

attrition rate from 2005/2006 to 2009/2010 was 15.7% for households and 20.5% for

individuals. For the following years, no explicit calculation for the attrition rate was offered

from LSMS and UNPS. By observing the original households from 2009/2010 to the

subsequent years, the calculation can be made. From the base year to 2010/2011, the attrition

rate is 11.02%, and from the base year to 2011/2012 is 13.04%, which is the overall attrition

rate. The construction of the survey weights provided from the LSMS UNPS survey for

Uganda account for several causes; an attrition correction factor to correct for households that

were unable to be surveyed and the incorporation of new households in the sample to account

for the changing population which is accomplished by adding split-off households to

recalibrate the sample size. The household panel weights are adjusted to correct for attrition

in each consecutive year using predicted response probabilities from a logistic regression

model. The methodology of the UNPS follows that of Rosenbaum and Ruben (1984). This

analysis uses the household probability weights with the intention of providing unbiased,

nationally representative parameter estimates for Uganda.

Incomplete Data:

Despite the stated sample size by the World Bank, of those reported figures, in 2009-2010

68.2% of the households reported the size of their landholdings. In 2010-2011, 72% reported

their landholding size, and in the 2011-2012 wave, 69.9% reported. Many of the calculations

make use of the reported household landholding sizes, making the sample used in estimations

a sub-sample. Aside from unreported landholding sizes, other variables have reporting

issues, as well, but not to the same extent as landholding size. Statistical software programs,

such as STATA 14 (used for all calculations in this analysis) that perform econometric

estimations, require all values of estimated variables to be reported if they are going to be

included in the estimation sample. Therefore, estimation samples may decline significantly

for such reasons. Missing values of variables can become a concerning issue because it can

be a source of estimate bias.


62

To account for any potential problems that could arise from the omitted observations in the

sample, imputed means are calculated for missing values of continuous variables to provide a

separate, parsimonious model. Although this method has its drawbacks, such as reduced

variance and can disrupt the correlation between variables, it is meant to provide a robustness

check on the original sample used for estimation. Tables of both estimations will be

provided.


63

CHAPTER 6: RESULTS

6.1 Bivariate Relationship Analysis

To foreshadow the analysis, bivariate analysis is provided on the Pearson linear correlation

coefficient between the explanatory variables and the dependent variables in each stage of the

triple-hurdle in table 6.1. This will aid in the interpretation of the results to come. Following

the triple hurdle estimation, average partial effects for the key variables of interest are

provided, along with a brief analysis of the common marketing channels Ugandan

smallholder farmers use for HVA products to provide greater context.

It should be noted that, in the estimations, the number of observations may decline severely

due to household observations with incomplete information. HVP represents the first stage

binary outcome of whether or not households are HVA producers. NS, Autarkic, NB denotes

the second stage ordered probit, determining whether households are net buyers, autarkic, or

net sellers in the market. Ln(Net Sales) and Ln(Net Purchases) represents net sales if the

household is a net seller and net purchases if the household is a net purchaser. In parentheses

are the t-statistics indicates if there is a statistically significant linear relationship between the

variables at an alpha level of 5%, which will be shown by an asterisk next to the correlation

coefficient.

Table 6.1: Correlations Between Dependent and Explanatory Variable

HVP NS, Autarkic, NB Ln(Net Sales) Ln(Net Purchases)

Male Head 0.077* 0.023 0.219* -0.030

(0.000) (0.055) (0.000) (0.518)

Ln(Household Land) 0.046* 0.048* 0.295* -0.160*

(0.000) (0.000) (0.000) (0.001)

Adult Equiv. 0.075* 0.002 0.177* 0.169*

(0.000) (0.882) (0.000) (0.000)

Ln(Prod. Assets) 0.105* 0.021 0.312* 0.247*

(0.000) (0.073) (0.000) (0.000)

Hired Labor 0.000 0.084* 0.167* 0.150*

(0.025) (0.000) (0.000) (0.001)

Improved Seed 0.051* 0.006 0.072* 0.006

(0.000) (0.668) (0.000) (0.896)

Tech Endowment 0.011 -0.035 0.053* 0.088

(0.410) (0.147) (0.001) (0.065)

Head Age 0.071* -0.033* -0.077* -0.035

(0.000) (0.004) (0.000) (0.462)

Head Education 0.009 0.000 0.099* 0.254*


64

(0.543) (0.978) (0.000) (0.000)

Received Credit -0.025* 0.070* 0.079* -0.038

(0.044) (0.000) (0.000) (0.413)

Urban -0.064* 0.011 -0.0261 0.183*

(0.000) (0.354) (0.090) (0.000)

Distance Pop Center -0.064* -0.037* 0.021 -0.168*

(0.000) (0.001) (0.175) (0.000)

Distance to Market -0.011 -0.085* 0.014 -0.155*

(0.384) (0.000) (0.353) (0.001)

Distance to Road 0.126* 0.046* 0.021 -0.157*

(0.000) (0.000) (0.173) (0.001)

Ann Avg. Rainfall 0.055* 0.115* -0.113* 0.089

(0.000) (0.000) (0.000) (0.057)

Elevation 0.055* 0.050* 0.077* 0.013

(0.000) (0.000) (0.000) (0.776)

Ln(dist Coff price) 0.108* -0.001 -0.022 0.207*

(0.000) (0.942) (0.258) (0.006)

Ln(dist Beans price) -0.062 0.008 -0.040* -0.028

(0.000) (0.584) (0.019) (0.652)

Ln(dist Maize price) 0.024 0.059* -0.026 0.091

(0.081) (0.000) (0.107) (0.076)

Ln(dist Cassava price) -0.031* 0.026 -0.020 -0.004

(0.041) (0.065) (0.263) (0.944)

Source: World Bank LSMS 2009/2010 n = 1,752

With the simple t-test, tech endowment, which indicates if the household has been exposed to

extension services, does not have a statistically significant linear relationship with households

that cultivate HVA. Most notably, land and productive assets ownership show significant

relationships in most if not all stages with the dependent variables. This result is not

unexpected, as these are both variables commonly theorized to increase with

commercialization. Surprisingly, distance to the nearest road has a significant, but positive

correlation with HVA producers, which is inconsistent with hypothesis. This would indicate

further access to roads is positively correlated to HVA producers. Contrarily, distance to the

nearest population center shows an inverse correlation with HVA producers, as expected.

This result implies closer distances to population centers increase with households that

cultivate HVA.

Common Marketing Channels of HVA:

To provide deeper context into Ugandan agricultural marketing, table 6.3 provides the

common marketing channels for HVA that farmers face. Government buyers and local

cooperatives have a relatively small presence in Uganda for HVA products, comprising less


65

than 1% of the marketing channels. Private traders in local markets take the lion’s share of

HVA purchases, indicating over half of the purchases, reaching 65.03% in 2009. Private

traders in the district market and the market consumer results are relatively balanced,

hovering around 9-20%.

Table 6.2: Marketing Channels of HVA

Year Gov/Local

Coop

Private Trader in

Local

Village/Market

Private

Trader in

Dist.

Market

Consumer

at Market

Neighbour/

Relative Other

2009-2010 0.61% 65.03% 11.54% 13.02% 6.54% 3.27%

2011-2011 0.41% 66.59% 9.37% 16.54% 6.13% 0.96%

2011-2012 0.11% 55.78% 17.32% 19.98% 6.23% 0.58%

Source: Uganda LSMS-UNHS survey data, n = 1,371 for 2009-2012 (In panel form).

Local neighbors or relatives are also marginal buyers of HVA products from producers, but

only comprises 6-7%. This finding stresses the importance of local markets. Their access

can be improved by reducing transaction costs, in ways such as increasing their proximity or

prevalence, road access, and ease of general transportation. Given the data availability of this

survey, it is not clear as to where HVA products are distributed beyond the marketing

channels mentioned above.

6.2 Triple Hurdle-Model

Results from the triple-hurdle estimation are presented in table 6.3. Time dummies for all but

the base year are included in estimations to allow for varying intercepts and coefficients

between time periods. To test for conditionally uncorrelated errors between the stages, the

IMR is generated around the probability of being an HVA producer and is included in the

second stage estimation as a regressor. Secondly, the IMR is then calculated from around the

stage two market participation outcomes, then included in stage three as a regressor. In both

cases, they are statistically insignificant, as seen below. The resulting coefficient and p-value

from a simple t-test to allow us to conclude there is no selection bias between the stages of

estimation. The coefficient of the IMR when included in the consequent second and third

stage estimations for net sales are the following:

Stage 2 Market Participation Outcome: 𝛽1𝐼𝑀𝑅1 = -3.944; p-value = 0.413

Stage 3 Net Sales: 𝛽2𝐼𝑀𝑅2 = 1.705; p-value = 0.367


66

Considering the lack of statistical significance of the Inverse Mills Ratio in explaining market

participation outcomes and net sales, the IMR is omitted from the estimates in the triple

hurdle.

The estimation for net purchases in the third stage, conditional on being a producer and net

buyer, has few observations, along with few statistically significant coefficients. The data

structure under the triple-hurdle indicates that, of those who produce HVA, they rarely result

in being a net buyer. Considering this, the estimation for net purchases is omitted from the

results because no meaningful inference can be made.

The results of the triple-hurdle model in table 6.3 are not marginal effects due to the non-

linear nature of the likelihood function used for estimation. The interpretation should

therefore be limited to the sign and statistical significance associated with the probability of

the results. For example, a positive coefficient for household land size in stage one indicates

greater likelihoods of being a producer of HVA are associated with increasing land sizes.

Stage two ordered probit coefficients can be interpreted similarly, with a small alteration;

greater probabilities of being a net seller are associated with increases in the coefficients,

while there are greater probabilities of being a net buyer as the coefficients decrease,

conditional on being a HVA producer (Burke et al. 2015). Stage three results provide

estimates for net sales quantities, conditional on being a HVA producer and net seller.

Beneath the coefficients are the standard errors. Asterisks indicate statistical significance at

different levels of alpha.

Table 6. 3: Triple Hurdle Estimation

Explanatory Variables Stage 1: Production

Decision

Stage 2: Market

Participation

Stage 3: Net

Sales

Ln(Land Size) 0.211** 0.347** 0.692***

(-0.0823) (-0.136) (-0.112)

Head Sex -0.15 -0.475*** 0.268*

(-0.126) (-0.178) (-0.15)

Dist. % Rec Credit -2.210*** -1.097 0.112

(-0.787) (-1.14) (-0.866)

Ln(Productive Assets) 0.0684 0.0871 0.219***

(-0.0436) (-0.0663) (-0.0651)

Head Education -0.00165 -0.0146*** 0.00161

(-0.00281) (-0.00358) (-0.00424)

Urban -0.492*** 0.178 -0.115

(-0.188) (-0.217) (-0.202)


67

Dist. Pop. Center -0.00447 0.0111 -0.00763

(-0.00504) (-0.00813) (-0.00607)

Dist. Market 0.0153** 0.0164 0.0205**

(-0.0077) (-0.0128) (-0.0104)

Dist. Road 0.000562 -0.00870** -0.00357

(-0.00317) (-0.00415) (-0.00369)

Improved Seed 0.399*** 0.568*** 0.195

(-0.132) (-0.213) (-0.149)

Head Age 0.00477 0.00296 -0.0108***

(-0.00368) (-0.00542) (-0.00395)

Annual Avg. Rainfall -0.00116** -0.00145** -0.00262***

(-0.000467) (-0.000714) (-0.000578)

Elevation 0.000456 0.00103** 0.000446

(-0.000338) (-0.000476) (-0.000367)

Hired Labor 0.567*** -0.442 -0.227

(-0.171) (-0.303) (-0.231)

Adult Equivalents 0.039 -0.0719* 0.0124

(-0.0273) (-0.0386) (-0.0301)

Extension Service 0.269*** 0.0881 0.115

(-0.0916) (-0.158) (-0.115)

Ln(Cassava Price) -0.0318 -0.0038 -0.00581

(-0.0242) (-0.0437) (-0.0306)

Ln(Beans Price) -0.0433* -0.0284 0.0645**

(-0.0242) (-0.0433) (-0.0305)

Ln(Maize Price) 0.0582*** 0.047 -0.00292

(-0.0212) (-0.0399) (-0.0253)

Ln(Coffee Price) 0.0104 0.0117 -0.0192

(-0.023) (-0.0341) (-0.0284)

2010 Dummy 0.158 0.159 0.419***

(-0.112) (-0.205) (-0.152)

2011 Dummy 0.0961 0.376 0.764***

(-0.147) (-0.232) (-0.184)

Constant -0.493 12.22***

(-0.907) (-1.144)

Cutoff 1 -1.778

(-1.353)

Cutoff 2 -1.472 (-1.343)

Sigma 0.323***

(-0.698)

% Correctly Classified 68.51%

Observations 1,353 867 761

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1


68

As a robustness check, table 6.4 below provides estimations for a parsimonious model of the

triple hurdle where missing values have been calculated for their imputed means at the

household-level. First, log-productive assets becomes a highly statistically significant

determinant in the first stage decision in the parsimonious model, not being significant in the

original. This provides useful information because ln(productive assets) was one of the

variables with many missing values. Also, the coefficient for full time adult equivalents

becomes statistically significant in the parsimonious model, along with district cassava price

and district coffee, while not being in the first. However, district beans price loses statistical

significance in the parsimonious estimations. Most of the coefficients stay relatively similar

between the original and parsimonious model. As a method to check for robustness, most of

the findings in the original model can be confirmed as consistent. For most of the

interpretation, the original model will be used.

Table 6.4: Triple Hurdle Estimation (Parsimonious Model with Imputed Means)

Explanatory Variables Stage 1: Production

Decision

Stage 2: Market

Participation

Stage 3: Net

Sales

Ln(Land Size) 0.130** 0.377*** 0.659***

(-0.061) (-0.119) (-0.0939)

Head Sex 0.118 -0.139 0.342**

(-0.0921) (-0.129) (-0.137)

Dist. % Rec Credit -1.963*** -1.425* -0.214

(-0.58) (-0.854) (-0.775)

Ln(Productive Assets) 0.101*** 0.129** 0.252***

(-0.0365) (-0.058) (-0.053)

Head Education 1.83E-05 -0.0110*** 0.00616*

(-0.00219) (-0.00293) (-0.00349)

Urban -0.343** 0.0219 -0.178

(-0.144) (-0.163) (-0.203)

Dist. Pop. Center -0.00369 0.0111* -0.0109**

(-0.00402) (-0.00607) (-0.00545)

Dist. Market 0.0148** 0.00588 0.0108

(-0.00581) (-0.0098) (-0.00846)

Dist. Road -0.00345 -0.00593 -0.00065

(-0.0027) (-0.0036) (-0.00342)

Improved Seed 0.368*** 0.306** 0.0721

(-0.0954) (-0.153) (-0.111)

Head Age 0.004 -0.00139 -0.0131***

(-0.00295) (-0.00414) (-0.00346)

Annual Avg. Rainfall -0.000902*** -0.00110*** -0.00233***

(-0.0003) (-0.0004) (-0.000374)


69

Elevation 0.000299 0.000713** 5.71E-06

(-0.000217) (-0.0003) (-0.000313)

Hired Labor 0.541*** -0.353 -0.508**

(-0.136) (-0.267) (-0.21)

Adult Equivalents 0.0371* -0.0576* 0.0369

(-0.0219) (-0.0297) (-0.0285)

Extension Service 0.129* -0.0033 0.0471

(-0.0692) (-0.115) (-0.102)

Ln(Cassava Price) -0.0449** 0.0033 0.00412

(-0.0198) (-0.0322) (-0.0279)

Ln(Beans Price) -0.0252 -0.0325 0.0363

(-0.0171) (-0.0304) (-0.0262)

Ln(Maize Price) 0.0421*** 0.0152 -0.0128

(-0.0152) (-0.026) (-0.0252)

Ln(Coffee Price) 0.0364** 0.0209 -0.0059

(-0.0153) (-0.0224) (-0.021)

2010 Dummy 0.116 0.0209 0.175

(-0.0816) (-0.164) (-0.117)

2011 Dummy 0.178* 0.102 0.483***

(-0.0959) (-0.157) (-0.127)

Constant -1.147* 12.72***

(-0.696) (-0.97)

Cutoff 1 -1.557

(-1.019)

Cutoff 2 -1.18

(-1.017)

Sigma 1.390443***

(0.9319)

% Correctly Classified 63.83%

Observations 2,480 1,495 1,290


*** p<0.01, ** p<0.05, * p<0.1

In the original estimations from table 6.3, the results begin with the first stage decision of

whether or not to produce HVA. Many of the theorized variables influencing agricultural

household decisions – access to information, infrastructure, inputs, capital, and crop prices

prove to be statistically significant factors in determining the probability of being a producer.

Among the most important is household land size. Komarek (2010) states that in Uganda

larger land size gives farmers greater capacity to allocate land to non-consumption crops

without raising food security concerns. Land size increases are not only associated with

greater probabilities of being a producer of HVA, but are also significant in determining

whether a household is a net-selling market participant and conditional sales quantities. In


70

this estimation, the coefficient for household land size supports such theories, further

suggesting increasing household land size is conducive to commercial farming in Uganda.

Access to information by way of exposure to extension services is a significant factor

influencing the production of HVA. Despite its noteworthy impact in stage one, its influence

proves to have less of an effect in the subsequent stages in determining market participation

outcomes and total net sales, conditional on being a net seller. This result proves to be

consistent with the hypothesis and the available information that various policies in Uganda

have sought to promote small-scale commercialization with extension services being a key

tool to do so. The results show promising outcomes for Uganda’s extension services and its

positive effect on the probability of producing HVA. In Uganda, this has been flagged as an

area of critical importance by various literature and national government.

In the parsimonious model, a significant relationship exists between the household’s value of

owned productive assets, a proxy representing household wealth, and the probability of

producing HVA in all stages. Results indicate that not only do household asset positively

influence HVA production, but are also associated with being net sellers in the market and

greater margins of net sales. Endogeneity concerns are frequent when using this as an

explanatory variable, therefore, results should be interpreted skeptically because this may not

indicate causality. HVA production is widely cited to have high start-up costs and is capital

intensive, therefore factors mitigating up-front costs and improve working capacity should, in

theory, facilitate HVA adoption. Consistent with theory, is its positive relationship when

predicting the margin of net sales, conditional on the producing household being a net seller.

The dummy variable to indicate use of improved seed is a significant factor positively

influencing HVA cultivation and being a net seller in stage two. These results are consistent

with hypothesis that greater farm inputs will facilitate HVA production.

Households obtaining credit at the district level provides interesting results in its relationship

with the first-stage decision to cultivate HVA. Results show farmers with greater shares of

district-level credit availability are less likely to cultivate HVA with a relatively large

magnitude. Immink and Alcaron (1993) stated low levels of credit availability have been

identified as a key constraint to technology adoption entry to cash crop markets. However,

the results obtained are mixed and inconsistent with expectations and prior theory. One


71

possible explanation may be due to the low development of credit markets in Uganda and

their access challenges, which has been cited on a variety of accounts5.

Increasing age of the household head shows no relationship in the first two stages. However,

its stage three outcome shows to be inconsistent with much of theory, showing a negative

coefficient, indicating that of HVA net sellers, increasing age, a proxy for experience, is

associated with declining net sales margins.

Gender of the household head shows interesting results; there is no significant relationship

between HVA production and male and female heads. However, male heads are more likely

to be net buyers in the market place if they produce HVA, whereas if the household is a net

seller, conditional on being a producer, male heads are more likely to have a greater net sales

margin than female heads. Years of education received from the household head shows no

relationship in the first stage, although we find, conditional on being a producer, increasing

years of education make the household more likely to be a net purchasing market participant.

A possible explanation for this is that farming is a secondary profession of those with greater

levels of education.

Full-time adult equivalents, (FTAE) which are an indicator of household labor availability,

does not show a positively relationship with the probability of producing HVA, while

producers of HVA are more likely to be net buyers as household’s FTAE increase. These

results in stage one are inconsistent with the notion that producers are more likely to cultivate

HVA with greater working capacity. Hired labor, however, which is also a measure of labor

availability is highly significant in influencing households to produce HVA. This may give

indication that, households that can afford to purchase inputs, are more apt to commercialize.

Infrastructural characteristics provide mixed results. Counterintuitively, further distances to

nearest market increase the probability of adopting HVA and the other infrastructure

variables are not statistically significant. Consistent with theory, urban households are less

likely to cultivate HVA. However, conditional on producing, increasing distances to the

nearest road are associated with being a net buyer in the market. Findings for annual average

rainfall are significant in all stages. In stage one, greater rainfall is associated with the

5 Munyambonera, E., Mayanja, M. L., Nampeewo, D., & Adong, A. (2014). Access and use of credit in Uganda:

Unlocking the dilemma of financing small holder farmers. Mpuga, P. (2010). Constraints in access to and

demand for rural credit: Evidence from Uganda. African Development Review, 22(1), 115-148.


72

likelihood of being not producing HVA. In stage two, households are more likely to be net

purchasers with greater levels of rainfall and lead to lower levels of net sales in stage three –

all which are inconsistent with the hypotheses.

Prices prove to be a significant determinant in the first stage decision to produce HVA.

Maize, beans and cassava, predominantly consumption crops, are hypothesized substitutes for

HVA products in Uganda. The price of beans appears to reduce the probability of cultivating

HVA with exception of maize, which increases the probability of HVA cultivation. This

outcome may be due to the fact many farmers will cultivate maize regardless of HVA

cultivation. Cassava shows no relationship in the decision to cultivate HVA. Increases in the

price of beans positively increase the net sales margin, conditional on the household

producing HVA and being a net seller.

Average Partial Effects and Probabilistic Outcomes:

This research hypothesizes smallholders that cultivate HVA are more likely to be net sellers

in the market. The probability of any given household being a producer of HVA is 39.3%.

The predicted probability that any given household that produces HVA is a net seller is

90.6% with a standard deviation of 0.073 using 1,353 observations. Furthermore, the

probability of any given HVA producer is a net buyer in the market is 5.6% (SD = 0.051),

while the probability they do not participate at all (autarkic) is 3.7% (SD = 0.0022). These

calculations provide unequivocal evidence that smallholders who cultivate HVA are highly

likely to orient themselves toward the market.

Burke et al. (2015) applies average partial effects (APE) to dichotomous explanatory

variables that indicate the presence of various marketing channels. These calculations

include the estimated partial effect on the probability of being a net selling producer, as well

as the estimated partial effect of unconditional expected values of net sales, with respect to

the presence of various dairy marketing channels. Given the wide range of HVA products in

this analysis, no such explanatory variables were included in this analysis. Therefore, rather

than use Burke’s approach, the APE are applied to key explanatory variables to observe how

they change unconditional expected values of net sales and the probability of being a net

selling producer. The selected explanatory variables include those that have statistical

significance and greater relevance to triple-hurdle study around HVA adoption and market

participation. These include extension service contact, land holding size, distance to market,


73

head age, urban versus rural households, use of improved seed, and productive assets.

Standard errors6 and the estimate’s statistical significance levels are obtained by

bootstrapping 100 replications. Table 6.5 provides this output of the estimated partial effect

on unconditional expected values of net sales. Following table 6.5 will show the results for

the APE on the probability of being a net selling producer across the selected explanatory

variables.

Table 6.5: Estimated Partial Effect on the Unconditional Expected Value of Net Sales

Column (1)

Explanatory Variables

Column(2)

APE on Unconditional Expected Value of

Net Sales (UGX) 𝜕(𝑛𝑒𝑡 𝑠𝑎𝑙𝑒𝑠|𝒙)

𝜕𝐿𝑛(𝐻𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑 𝐿𝑎𝑛𝑑) 279,010.3**

(123111.9)

𝜕(𝑛𝑒𝑡 𝑠𝑎𝑙𝑒𝑠|𝒙)

𝜕(𝐸𝑥𝑡𝑒𝑛𝑠𝑖𝑜𝑛 𝑆𝑒𝑟𝑣𝑖𝑐𝑒) 73,294.41

(64654.28)


𝜕(𝐷𝑖𝑠𝑡𝑎𝑛𝑐𝑒 𝑡𝑜 𝑀𝑎𝑟𝑘𝑒𝑡) -6,257.88*

(-5494.97)


𝜕𝐿𝑛(𝑃𝑟𝑜𝑑𝑢𝑐𝑡𝑖𝑣𝑒 𝐴𝑠𝑠𝑒𝑡𝑠) 72,284.25

(59663.02)


𝜕(𝐻𝑒𝑎𝑑 𝐴𝑔𝑒) 1731.14

(5303.20)


𝜕(𝑈𝑟𝑏𝑎𝑛) 108,393.2

(156594.5)


𝜕(𝐼𝑚𝑝𝑟𝑜𝑣𝑒𝑑 𝑆𝑒𝑒𝑑) 417214.5**

(169415.9)


*** p<0.01, ** p<0.05, * p<0.1

The results in table 6.5 are an important part of this study because these are the factors that

will improve marketing opportunities for households that produce HVA. The results of the

APE on the unconditional expected value of net sales provide indication that of HVA

producing households, increases in landholdings, lower transfer costs, and use of improved

seed increase the margin of net sales. A one standard deviation change in household land size

is associated with a (UGX) 279,010.3 increase in expected net sales. Additionally, ability to

6 Robust standard errors of the APE on the expected quantity of net sales and the probability of being a net

selling producer of HVA are computed using statistical bootstrapping that relies on random sample replacement

of 100 repetitions.


74

access local markets in terms of market distance supports theory. With every kilometer

further away from the market that a household sits, one can expect a decrease of (UGX)

6,257.88 in market sales with an alpha level of 0.10. With respect to households that are

exposed to extension services, the coefficient of the APE was not statistically significant,

though the effect shows an average increase of approximately (UGX) 73,294.41. The APE

for log productive assets are not statistically significant in explaining unconditional expected

net sales. Uganda is densely populated and theory suggests that urban households have the

potential to show negative returns to the market for a variety of reasons, however the APE for

this variable is not statistically significant. HVA producing households that use improved

seed are expected to have an average of 417,214.5 greater net sales in the market than those

that do not. Following in table 16 shows the APE on the probability of being a net selling

producer.

Table 6.6: Estimated Average Partial Effect on the Probability of being a Net Selling

Producer of HVA

Column (1)

Explanatory Variables

Column(2)

APE on Probability of being a Net Selling

Producer 𝜕𝑃𝑟(𝑁𝑒𝑡 𝑆𝑒𝑙𝑙𝑒𝑟|𝒙)

𝜕𝐿𝑛(𝐻𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑 𝐿𝑎𝑛𝑑) -.025

(0.920)

𝜕𝑃𝑟(𝑁𝑒𝑡 𝑆𝑒𝑙𝑙𝑒𝑟|𝒙)

𝜕(𝐸𝑥𝑡𝑒𝑛𝑠𝑖𝑜𝑛 𝑆𝑒𝑟𝑣𝑖𝑐𝑒) -.517

(0.868)


𝜕(𝐷𝑖𝑠𝑡𝑎𝑛𝑐𝑒 𝑡𝑜 𝑀𝑎𝑟𝑘𝑒𝑡) -.005

(0.784)


𝜕𝐿𝑛(𝑃𝑟𝑜𝑑𝑢𝑐𝑡𝑖𝑣𝑒 𝐴𝑠𝑠𝑒𝑡𝑠) -0.096

(0.812)


𝜕(𝐻𝑒𝑎𝑑 𝐴𝑔𝑒) -.008

(0.861)


𝜕(𝑈𝑟𝑏𝑎𝑛) 1.130

(0.764)


𝜕(𝐼𝑚𝑝𝑟𝑜𝑣𝑒𝑑 𝑆𝑒𝑒𝑑) -.527

(0.901)


*** p<0.01, ** p<0.05, * p<0.1


75

As opposed to the APE of the selected explanatory variables on the expected quantity of net

sales, which appear intuitive and match theory, their APE on the probability of being a net-

selling producer gives mixed results. Several of their signs do not match the expected signs

put forward in table 4.1. Access to information via extension services, shows to have a

negative probability of being a net seller if the household produces. Despite many of the

coefficients here not being statistically significant, this can be seen as a valuable finding in

itself. One conclusion that can be possibly be made is the variables that influence the market

participation outcomes are different from those that influence a given household to cultivate

HVA. This conclusion can be made based on the statistical significance found in many of the

stage one results, whereas, their APE on the probability of being a net seller prove to be

statistically insignificant.

Increased agricultural productivity via improving sales margins plays a key role in

transforming rural agriculture. Shifting resources away from staple crops into HVA has

proven to be one method to improve rural incomes. HVA producers are more likely to be net

sellers in markets, demonstrating how smallholder crop diversification into higher valued

products can lead to agricultural transformation.

The first hypothesis of this research can to a large degree be validated by observing the

statistical significance of the variables that increase the probability of a household cultivating

HVA in the stage one estimates. As stated, access to information, infrastructure, inputs,

capital, and crop prices, are significant determinants of HVA cultivation. The parsimonious

model shows that the value of household productive assets is a statistically significant

determinant of cultivating HVA. This may information may indicate the importance of

capital in the decision to cultivate HVA. Thus, if policy makers wish to better understand the

dynamics driving smallholder adoption of higher valued crops, they must better understand

these factors and their interplay to facilitate this process or make projections about cropping

patterns.

It appears that, inherently, households who produce HVA are more likely to be net sellers in

the market than they are to be autarkic or net buyers – based on estimated predicted

probabilities. Which of the selected variables change the probability of a household being a

net seller of the HVA producers, could not be explained through the APE with statistical

significance – inconsistent with hypothesis. However, several of these same factors do

determine market participation outcomes with statistical significance, though their precise


76

probability cannot be inferred. It can be concluded from the triple hurdle model framework

that households who produce HVA are more likely to be net sellers than not, but their

specific probability lacks significance, although many of their signs match predictions with

statistical significance. Thus, policies facilitating HVA adoption will likely result in a greater

quantity of net selling smallholder farmers in the population.

6.3 Policy Recommendations

The findings of this research hold numerous policy implications designed for strategies that

will promote agricultural transformation by creating a conducive environment and

encouraging farmers to cultivate HVA which can lead to greater net sales margins.

Suggestions will be made based on Uganda-specific literature and the findings of this study.

• If governments wish to promote greater market opportunities for smallholder

farmers in Uganda, policies should focus specifically on strategies to increase its

uptake. Those who make the decision to cultivate are not only likely to be more

commercially oriented market participants, but also more likely to be net sellers rather

than net purchasers in the market.

• Creating an environment where smallholders have access to land allowing them

to bring greater quantities of land under cultivation will promote HVA

production and greater market sales. Strengthening the security of land tenure and

implementing a more flexible land market system are cited factors Uganda needs to

focus on for smallholder land acquisitions and greater investments in HVA (Holden et

al. 2014). From a food security perspective, there is broad agreement that the

inclusion of HVA or non-food crops does not compete with crops produced for home

consumption in Uganda (Komarek 2010).

• To promote HVA production, improving the quality and capacity of extension

services to facilitate technology transfers via technical training on HVA practices

and marketing opportunities will be required. Evidence shows that exposure to

extension services and farming experience play an important role in HVA uptake

decisions. However, Uganda has faced routine difficulties with agricultural extension

services (FAO 2016).

• Infrastructural improvements, specifically ones that facilitate market access,

should be stressed. Considering that the bulk of HVA sales are made in local village

markets and the significance of market distance in increasing sales margins for HVA


77

producers, improving road quality of public infrastructure and strengthening the

linkage between market channels and rural farmers needs to be carried out.

Additionally, strategic investments in infrastructure specific areas, such as transport,

support for farming organizations, agricultural extension facilities will integrate

farmers better into supply chains and markets.

• Capacitating less wealthy farmers with working capital and inputs will result in

greater quantities of farmers producing HVA improved marketing

opportunities, along with increased net sales for those who produce. Increasing

levels of wealth and inputs, such as improved seed and hired labor are associated with

greater probabilities of HVA production and increased net sales margins. Evidence

shows that capital and inputs are key factors to facilitate adoption and greater sales

margins. The implications of this result, is that greater levels of working capital will

assist in reaching HVA production objectives.

• Capacitating women, relieving social barriers and improving their access to

education are cited strategies to improve gender differences in farming outcomes

(Farnworth, Akamandisa, and Hichaambwa 2011). Differences in gender with

respect to net sales among HVA producers warrant significant attention. Policies with

this orientation will decrease gaps in marketing outcomes between male and female

heads, specifically for HVA producers. Male headed households have approximately

27% greater net sales margins than female headed households that produce HVA.

• Consistency with farm gate prices will decrease farmers expected risk when

cultivating new crops, such as HVA, therefore policies to stabilize price are

integral. Uganda has been characterized as having large price fluctuations with

constistently volatile markets (Komarek 2010). Ceteris paribus, prices influence

Ugandan farmers’ decisions to cultivate HVA.

Some research has speculated that policies that facilitate market participation of a set of

products, such as HVA may also induce their cultivation. In Uganda, it is difficult to isolate

which factors contribute to both. However, those who do cultivate HVA are more likely to

have improved marketing outcomes, thus policies that focus on its uptake should be of focus.


78

CHAPTER 7: CONCLUSIONS

Theories of agricultural transformation posit that factor shifts from agriculture to non-

agricultural sectors play an important role in economic growth and poverty reduction (Shilpi

et al. 2016). Governments have attempted to devise strategies to identify what key factors

catalyze these processes of structural transformation. This research argues that crop

diversification can play an important role in this process. Therefore, advancing our

understanding of the factors associated with decisions made by smallholder farmers around

which crops to cultivate is critical to realizing these goals.

Historically, staple grains have dominated African agriculture due to food security needs, low

levels of rural development, and subsidies that have promoted a narrow range of crops.

Evidence is becoming abundantly clear that diversified products away from staple grains,

needs to occur for greater growth in the agricultural sector. In Uganda, like many other sub-

Saharan African countries, significant demand has emerged in regional, as well as

international markets, for high value agricultural products. This is the result of increased

incomes, urbanization and integrated markets. Yet, the share of high-value agriculture in

total exports out of sub-Saharan Africa remains at a low level. A significant opportunity

remains for sub-Saharan Africa, especially its smallholder farmers, to capitalize in supplying

these global markets for high-value products. This research analyzes what factors can

influence a smallholder farmers’ decision to commence producing HVA and how that can

lead to their different marketing outcomes and income generation.

Borrowing from Burke et al. (2015) this research uses a triple-hurdle model, which offers a

new methodology that gives broader insights into market participation decisions and

outcomes than before, involving an analysis of sequential decisions made by smallholder

farmers. Such analyses are appropriate for less commonly produced crops, such as HVA or

dairy.

The rationale behind using a triple-hurdle approach is to incorporate a third-stage involving

the question to first cultivate HVA, which can have specific policy relevance. Prior models

based on market participation use two-stage decision models, such as the double-hurdle,

which allows researchers to formulate conclusions based on the selected factors associated

with market participation along with its extent, in the second stage. However, policies

oriented toward market participation may be underestimating the impacts they have on


79

smallholder farming decisions. For example, policies that promote market participation for a

particular product, may also be inducing non-producers of HVA to commence producing.

Such policies can therefore have further reaching impacts than ones thought to only facilitate

market participation. The triple hurdle model seeks to understand exactly this, could market

participation strategies also be incentivizing smallholders to cultivate specific goods? These

results can be used to guide policy that will encourage HVA cultivation.

Panel data from the World Bank’s LSMS/UNHS surveys of Uganda is used for study in this

research. While controlling for the factors that are hypothesized to influence crop diversity

into higher valued products, triple hurdle results show that access to information,

infrastructure, head education, rural households, inputs such as adult equivalents and

improved seed use, capital, and crop prices, are significant determinants of HVA cultivation.

A parsimonious model was also estimated using imputed means of variables that had missing

values. In this estimation, the value of productive assets became highly statistically

significant, which provides additional information pointing to the importance of capital in the

decision to cultivate HVA. Additionally, results indicate that households that produce HVA

are 90% more likely to be net sellers in the market than they are to be autarkic or net buyers –

estimated based on the predicted probabilities. This could not be explained through isolating

which factors contribute to greater probabilities of being a net selling household with average

partial effects of the selected variables due to a lack of statistical significance. However,

stage two estimates do suggest that household land size, head education, distance to market,

hired labor and adult equivalents all play a significant role in determining a household’s

marketing outcomes, of HVA producers, be it net buyers, autarkic or net sellers.

Considering the array of initiatives carried out in Uganda to promote commercialization and

crop diversification, extension services who can relay these national objectives to smallholder

farmers are a key variable of study. Results provide valid indication that they are a strong

driver of HVA cultivation with the largest magnitude on the probability of being a producer

of HVA of all factors.

Infrastructure in Uganda has demonstrated to be a key area in need of policy focus. Public

investments must be made in infrastructural facilities, particularly those that contribute to the

ease of market access. Approximately 60-65% of smallholder HVA producers sell their

produce in local village markets. Dually, the distance to nearest market and population center

plays an important role in their market involvement and decision to cultivate. Therefore,


80

focus on transport, support for farming organizations, agricultural extension facilities will

integrate farmers better into supply chains and markets. Furthermore, gender, productive

assets, land ownership and prices proved to be among the most important factors surrounding

HVA marketing and its outcomes.

Considering the results and literature reviewed in this study, it is clear that improving

smallholder marketing outcomes through greater incomes can be achieved by promoting

HVA adoption. However, this is unlikely to be realized without effective policy

implementation that creates an enabling environment for smallholder farmers to do so.

Access to information, specifically through extension services that can transfer technology

and knowledge around HVA, improved infrastructure that facilitates market access and

lowers transaction costs, price stability, as well as input accessibility that improve working

capacity, are all meaningful ways of working toward this objective. Most importantly, policy

makers need to recognize that these factors are dynamic and do not exist or act in a closed

system. Understanding their intersection and depth will help guide strategies for sustainable

outcomes.


81

BIBILOGRAPHY

Abro, A.A., 2012. Determinants of crop diversification towards high value crops in Pakistan. International

Journal of Business Management & Economic Research, 3(3), pp.536-545.

AFRICA AGRICULTURE STATUS REPORT: Progress Toward Agricultural Transformation in Africa.

Growing Africa’s Agriculture, (2016). [online] Available at:

http://fsg.afre.msu.edu/outreach/2016AfricaAgriculturalStatusReport.pdf [Accessed 7 Jul. 2017].

Agriculture and small scale producers in Uganda: Issues, challenges and options for policy. Development

Research and Training, (2012). [online] Available at: http://www.drt-ug.org/wp-

content/uploads/2016/11/Agriculture-and-small-scale-producers-in-Uganda.pdf [Accessed 9 Sep. 2017].

Ahmadzai, H., 2017. Crop Diversification and Technical Efficiency in Afghanistan: Stochastic Frontier

Analysis (No. 2017-04).

Ahn, C.Y., Singh, I. and Squire, L., 1981. A model of an agricultural household in a multi-crop economy: the

case of Korea. The Review of EcoSingh, I., Squire, L. and Kirchner, J., 1985. Agricultural pricing and marketing

policies in an African context: a framework for analysis. The World Bank.

Alvarez-Cuadrado, F. and Poschke, M., 2011. Structural change out of agriculture: Labor push versus labor

pull. American Economic Journal: Macroeconomics, 3(3), pp.127-158.

Anderman, T.L., Remans, R., Wood, S.A., DeRosa, K. and DeFries, R.S., 2014. Synergies and tradeoffs

between cash crop production and food security: a case study in rural Ghana. Food security, 6(4), pp.541-554.

Ariga, J., Jayne, T.S. and Nyoro, J., 2008. Tegemeo Institute of Agricultural Policy and Development.

Barnum, H.N. and Squire, L., 1980. A model of an agricultural household. Theory and evidence (No. 27).

Barrett, C.B., Christiaensen, L., Sheahan, M. and Shimeles, A., 2017. On the structural transformation of rural

Africa. Journal of African Economies, 26(suppl_1), pp.i11-i35.

Barungi, M., Guloba, M. and Adong, A., 2016. Understanding the rice value chain in Uganda-opportunities

and challenges to increased production (No. 253558).

Bellemare, M., 2012. The inverse farm size–productivity relationship:“Proof” that smallholders can feed the

world. Accessed November, 3, p.2015.

Bellemare, M.F. and Barrett, C.B., 2006. An ordered Tobit model of market participation: Evidence from Kenya

and Ethiopia. American journal of agricultural economics, 88(2), pp.324-337.

Benin, S., Smale, M., Gebremedhin, B., Pender, J. and Ehui, S.K., 2003. THE DETERMINANTS OF CEREAL

CROP DIVERSITY ON FARMS IN THE ETHIOPIAN HIGHLANDS/доклад на 25 конференции IAAE,

Reshaping Agriculture’s Contribution to Society, International Convention Centre, Durban, South Africa, 16-23

August 2003.

Benin, S., Smale, M., Pender, J., Gebremedhin, B. and Ehui, S., 2004. The economic determinants of cereal crop

diversity on farms in the Ethiopian highlands. Agricultural Economics, 31(2‐3), pp.197-208.

Bennett, M.K., 1954. The World's Food (Vol. 78, No. 1, p. 83). LWW.

Berendsen, B., Dietz, T., Nordholt, H.S. and van der Veen, R., 2013. Asian tigers, African lions: comparing the

development performance of Southeast Asia and Africa. Brill.

Bezabih, M. and Sarr, M., 2012. Risk preferences and environmental uncertainty: Implications for crop

diversification decisions in Ethiopia. Environmental and Resource Economics, pp.1-23.

Birthal, P.S., Joshi, P.K., Roy, D. and Thorat, A., 2007. Diversification in Indian Agriculture Towards High-

value Crops (Vol. 727). Intl Food Policy Res Inst.


82

Bureau for Food and Agricultural Policy, 2016. Agricultural Outlook 2016-2025. pp. 148-149.

Burke, W.J., Myers, R.J. and Jayne, T.S., 2015. A triple-hurdle model of production and market participation in

Kenya's dairy market. American Journal of Agricultural Economics, 97(4), pp.1227-1246.

Carter, M.R., 2000. Land ownership inequality and the income distribution consequences of economic

growth (No. 201).

Chand, R., 1995. Agricultural Diversification and Small Farm Development in Western Himalayan Region.

In National Workshop on Small Farm Diversification: Problems and Prospects", NCAP, New Delhi.

Chaplin, H., 2000. Agricultural diversification: a review of methodological approaches and empirical

evidence. Online: http://www. ilr. uni-bonn.de/agpo/rsrch/Idara/Farm/wyewp2. doc (accessed 22 January

2013).

Chatterjee, S. and Wernerfelt, B., 1991. The link between resources and type of diversification: Theory and

evidence. Strategic management journal, 12(1), pp.33-48.

Chauvin, N.D., Porto, G. and Mulangu, F., 2017. Agricultural Supply Chains, Growth and Poverty in Sub-

Saharan Africa: Market Structure, Farm Constraints and Grass-root Institutions. Springer.

Chavas, J.P., 2001. An international analysis of agricultural productivity. FAO ECONOMIC AND SOCIAL

DEVELOPMENT PAPER, pp.21-38.

Chayanov, A.V. and Chayanov, A.V., 1925. The theory of peasant economy.

Christensen, L.R., Jorgenson, D.W. and Lau, L.J., 1975. Transcendental logarithmic utility functions. The

American Economic Review, 65(3), pp.367-383.

Christiaensen, L., Weerdt, J. and Todo, Y., 2013. Urbanization and poverty reduction: the role of rural

diversification and secondary towns. Agricultural Economics, 44(4-5), pp.435-447.

Cia.gov. (2017). The World Factbook — Central Intelligence Agency. [online] Available at:

https://www.cia.gov/library/publications/the-world-factbook/index.html [Accessed 30 Jul. 2017].

Cragg, J.G., 1971. Some statistical models for limited dependent variables with application to the demand for

durable goods. Econometrica: Journal of the Econometric Society, pp.829-844.

Davis, J.R., 2006. How can the poor benefit from the growing markets for high value agricultural products?

Dawe, D., 2015. Agricultural transformation of middle-income Asian economies: diversification, farm size and

mechanization.

De Janvry, A., Fafchamps, M. and Sadoulet, E., 1991. Peasant household behaviour with missing markets: some

paradoxes explained. The Economic Journal, 101(409), pp.1400-1417.

De, U.K., 2005. Economics of Crop Diversification–An Analysis of Land Allocation towards Different Crops .

Delgado, C., Hazell, P., Hopkins, J. and Kelly, V., 1994. Promoting intersectoral growth linkages in rural Africa

through agricultural technology and policy reform. American Journal of Agricultural Economics, 76(5),

pp.1166-1171.

Dercon, S., 1996. Risk, crop choice, and savings: Evidence from Tanzania. Economic development and cultural

change, 44(3), pp.485-513.

Di Falco, S. and Chavas, J.P., 2009. On crop biodiversity, risk exposure, and food security in the highlands of

Ethiopia. American Journal of Agricultural Economics, 91(3), pp.599-611.

Dorosh, P. and Thurlow, J., 2009. Agglomeration, migration, and regional growth: A CGE analysis for

Uganda (Vol. 848). Intl Food Policy Res Inst.

Dunkley, S. (2017). [online] Available at: https://www.linkedin.com/pulse/ugandas-declining-coffee-market-its-

effect-national-poverty-dunkley [Accessed 11 Jul. 2017].


83

Emran, M. Shahe, and Forhad Shilpi. "The extent of the market and stages of agricultural

specialization." Canadian Journal of Economics/Revue canadienne d'économique 45.3 (2012): 1125-1153.

Evans, N.J. and Llbery, B.W., 1993. The pluriactivity, part-time farming, and farm diversification

debate. Environment and Planning A, 25(7), pp.945-959.

Fafchamps, M., 1992. Cash crop production, food price volatility, and rural market integration in the third

world. American journal of agricultural economics, 74(1), pp.90-99.

FAO (2017). FAOSTAT database collections. Food and Agriculture Organization of the United Nations. Rome.

Access date: 2017-08-22. URL: http://FAOstat.FAO.org

Farnworth, C.R., Akamandisa, V.M., Hichaambwa, M. and Cornwall, P., 2011. Zambia feed the future gender

assessment. Lusaka, Zambia: United States Agency for International Development.

Fermont, A.V., Van Asten, P.J.A. and Giller, K.E., 2008. Increasing land pressure in East Africa: the changing

role of cassava and consequences for sustainability of farming systems. Agriculture, ecosystems &

environment, 128(4), pp.239-250.

Food and Agricultural Organization of the UN, (2017). [online] Available at:

http://www.FAO.org/docrep/013/i2050e/i2050e01.pdf [Accessed 11 Jul. 2017].

Gehlhar, M. and Regmi, A., 2005. Factors shaping global food markets. New directions in global food markets,

pp.5-17.

Gerold, G. and Fremerey, M., 2004. Land use, nature conservation and the stability of rainforest margins in

southeast Asia. Springer Science & Business Media.

Ghosh, M., Sarkar, D. and Roy, B.C. eds., 2014. Diversification of Agriculture in Eastern India. Springer.

Goetz, S.J., 1992. A selectivity model of household food marketing behavior in sub-Saharan Africa. American

Journal of Agricultural Economics, 74(2), pp.444-452.

Gollin, D., Parente, S. and Rogerson, R., 2002. The role of agriculture in development. The American Economic

Review, 92(2), pp.160-164.

Government Citizen Interaction Centre(GCIC). (2016). Government Citizens Interaction Centre -GCIC | Open

Government. [online] Available at: http://gcic.gou.go.ug/ [Accessed 16 Jul. 2017].

Haggblade, S. and Dewina, R., 2010, January. Staple food prices in Uganda. In Prepared for the Comesa policy

seminar on “Variation in staple food prices: causes, consequence, and policy options,” Maputo,

Mozambique (pp. 25-26).

Haggblade, S., Hazell, P. and Reardon, T., 2010. The rural non-farm economy: Prospects for growth and

poverty reduction. World Development, 38(10), pp.1429-1441

Hammel, E.A., 2005. Chayanov revisited: A model for the economics of complex kin units. Proceedings of the

National Academy of Sciences of the United States of America, 102(19), pp.7043-7046.

Holden, S.T. and Otsuka, K., 2014. The roles of land tenure reforms and land markets in the context of

population growth and land use intensification in Africa. Food Policy, 48, pp.88-97.

Holloway, G.J., Barrett, C.B. and Ehui, S.K., 2005. Bayesian estimation of the double hurdle model in the

presence of fixed costs.

Ibrahim, H., Rahman, S.A., Envulus, E.E. and Oyewole, S.O., 2009. Income and crop diversification among

farming households in a rural area of north central Nigeria. Journal of Tropical Agriculture, Food, Environment

and Extension, 8(2), pp.84-89.

Immink, M.D. and Alarcon, J.A., 1993. Household income, food availability, and commercial crop production

by smallholder farmers in the western highlands of Guatemala. Economic Development and Cultural

Change, 41(2), pp.319-342.


84

INCLUSIVE GROWTH POLICY NOTE 2 AGRICULTURE FOR INCLUSIVE GROWTH IN UGANDA.

World Bank, (2016). [online] Available at:

http://siteresources.worldbank.org/INTDEBTDEPT/Resources/468980-

1316457581843/CaseStudy_Uganda_01.pdf [Accessed 29 Jul. 2017].

Jaleta, M., Gebremedhin, B. and Hoekstra, D., 2009. Smallholder commercialization: Processes, determinants

and impact.

Johnston, B.F. and Mellor, J.W., 1961. The role of agriculture in economic development. The American

Economic Review, 51(4), pp.566-593.

Kabeere, F. and Wulff, E., 2008. Seed sector country profile: Uganda. Volume I: Overview of seed supply

systems and seed health issues. Copenhagen: Danish Seed Health Centre for Developing Countries, University

of Copenhagen.K

Katchova, A.L. and Miranda, M.J., 2004. Two-step econometric estimation of farm characteristics affecting

marketing contract decisions. American Journal of Agricultural Economics, 86(1), pp.88-102.

Kibaara, B., Ariga, J., Olwande, J. and Jayne, T.S., 2009. Trends in Kenyan agricultural productivity: 1997-

2007. Egerton university. Tegemeo institute of agricultural policy and development.

Komarek, A.M., 2010. Crop diversification decisions: the case of vanilla in Uganda. Quarterly Journal of

International Agriculture, 49(3), pp.227-242.

Kosgei (2013). News | FARMD: Forum for Agricultural Risk Management in Development. [online]

Agriskmanagementforum.org. Available at: http://www.agriskmanagementforum.org/news/kenya-government-

support-agricultural-diversification. [Accessed 19 Jul. 2017]. (Kosgei 2012)

Koutchade, P., Féménia, F. and Carpentier, A., 2014. Accounting for unobserved heterogeneity in agricultural

production choice models: a random parameter approach. In Annual Meeting of the Agricultural and Applied

Economics Association.KRISHI 2011

Kumar, A., Kumar, P. and Sharma, A.N., 2012. Crop diversification in Eastern India: Status and

determinants. Indian Journal of Agricultural Economics, 67(4), pp.600-616.

Kuroda, Y. and Yotopoulos, P.A., 1978. A Microeconomic Analysis of Production Behavior of the Farm

Household in Japan‐A Profit Function Approach, 29(2), pp.116-129.

Kurosaki, T., 2003. Specialization and diversification in agricultural transformation: the case of West Punjab,

1903–92. American Journal of Agricultural Economics, 85(2), pp.372-386.

Kuznets, S., 1973. Modern economic growth: findings and reflections. The American economic review, 63(3),

pp.247-258.

Leliveld, A.H.M., Dietz, A.J., Foeken, D.W.J., Klaver, W., Akinyoade, A., Smits, H., Soeters, S. and Wout, M.,

2013. Agricultural dynamics and food security trends in Uganda.

Lewis, W.A., 1954. Economic development with unlimited supplies of labour. The manchester school, 22(2),

pp.139-191.

Malabo Declaration on Accelerated Agricultural Growth and Transformation for Shared Prosperity and

Improved Livelihoods. African Union Commission, (2014). [online] Available at:

https://au.int/sites/default/files/documents/31006-doc-malabo_declaration_2014_11_26-.pdf [Accessed 17 Sep.

2017].

Mandal, R. and Bezbaruah, M.P., 2013. Diversification of cropping pattern: Its determinants and role in flood

affected agriculture of Assam Plains. Indian Journal of Agricultural Economics, 68(2), pp.170-181.

Masih, I., Maskey, S., Mussá, F.E.F. and Trambauer, P., 2014. A review of droughts on the African continent: a

geospatial and long-term perspective. Hydrology and Earth System Sciences, 18(9), p.3635.


85

Mather, D.L. and Jayne, T.S., 2011. The impact of state marketing board operations on smallholder behavior

and incomes: the case of Kenya. Michigan State University, Department of Agricultural, Food, and Resource

Economics and Department of Economics.

McGuire, R.A., 1980. A portfolio analysis of crop diversification and risk in the cotton south. Explorations in

Economic History, 17(4), pp.342-371.

Mellor, J.W., 1976. New economics of growth. Cornell University Press.

Mesfin, W., Fufa, B. and Haji, J., 2011. Pattern, trend and determinants of crop diversification: empirical

evidence from smallholders in eastern Ethiopia. Journal of Economics and Sustainable Development, 2(8),

pp.78-89.

Metcalf, D., 1969. The economics of agriculture. The economics of agriculture.

Minority Rights Group. (2011). Uganda - Minority Rights Group. [online] Available at:

http://minorityrights.org/country/uganda/ [Accessed 1 Sep. 2017].

Minot, N. ed., 2006. Income diversification and poverty in the Northern Uplands of Vietnam (Vol. 145). Intl

Food Policy Res Inst.

Mofya-Mukuka, R. and Hichaambwa, M., 2016. Factors influencing smallholder crop diversification in zambia

and the implications for policy. Working paper 112.

Moore, C.V. and Snyder, J.H., 1969. Crop Selection in High-Risk Agriculture. Agricultural Economics

Research, 21.

Mpuga, P. (2010). Constraints in access to and demand for rural credit: Evidence from Uganda. African

Development Review, 22(1), 115-148.

Mubiru, D.N., Komutunga, E., Agona, A., Apok, A. and Ngara, T., 2012. Characterising agrometeorological

climate risks and uncertainties: Crop production in Uganda. South African Journal of Science, 108(3-4), pp.108-

118.

Mukherjee, S., 2010. Crop diversification and risk: an empirical analysis of Indian states.

Mulumba, J.W., Nankya, R., Adokorach, J., Kiwuka, C., Fadda, C., De Santis, P. and Jarvis, D.I., 2012. A risk-

minimizing argument for traditional crop varietal diversity use to reduce pest and disease damage in agricultural

ecosystems of Uganda. Agriculture, ecosystems & environment, 157, pp.70-86.

Munyambonera, E., Mayanja, M. L., Nampeewo, D., & Adong, A. (2014). Access and use of credit in Uganda:

Unlocking the dilemma of financing small holder farmers.

Naads.or.ug. (2017). NAADS: National Agricultural Advisory Services. [online] Available at:

http://www.naads.or.ug/data/mreports/32/NAADS%20is%20working%20with%20Operation%20Wealth%20Cr

eation%20to%20transform%20the%20lives%20of%20Ugandans%20through%20improved%20Agricultural%2

0production...html [Accessed 1 Oct. 2017].

Nakajima, C., 1959. Subjective equilibrium theory of the farm household. (In Japanese) The Economic Review

of Osaka University, 7(2), No. 2

Nakajima, C., 1969. Subsistence and commercial family farms: some theoretical models of subjective

equilibrium. Subsistence agriculture and economic development, 165, p.185.

NASA (2000)Land Data Products and Services. [online] Available at:

https://lpdaac.usgs.gov/data_access/data_pool" [Accessed 6 Jul. 2017].

NEMA, (2010). [online] Available at: http://www.nemaug.org/reports/national_state_report_2010.pdf

[Accessed 1 Aug. 2017].

New Partnership for Africa’s Development, (2017). [online] Available at:

http://www.nepad.org/programme/comprehensive-africa-agriculture-development-programme-caadp. [Accessed

12 Aug. 2017].


86

Newby, H., 1988. The countryside in question. Hutchinson Education.

O'Neal, C., 2017. Do Farm Subsidies Affect Crop Diversification?.

Otsuka, K., Chuma, H. and Hayami, Y., 1992. Land and labor contracts in agrarian economies: theories and

facts. Journal of Economic Literature, 30(4), pp.1965-2018.

Quam, M.D., 1997. Creating peace in an armed society: Karamoja, Uganda, 1996. African Studies

Quarterly, 1(1), pp.33-46.

Ranis, G., Stewart, F. and Ramirez, A., 2000. Economic growth and human development. World

development, 28(2), pp.197-219.

Ranis, G., Stewart, F. and Reyes, E.A., 1990. Linkages in developing economies: A Philippine study (No. 1). Ics

Pr.

Rao, P.P., Birthal, P.S. and Joshi, P.K., 2006. Diversification towards high value agriculture: role of

urbanisation and infrastructure. Economic and Political Weekly, pp.2747-2753.

Ray, D., 2010. Uneven growth: A framework for research in development economics. The Journal of Economic

Perspectives, 24(3), pp.45-60.

Reardon, T. and Timmer, C.P., 2014. Five inter-linked transformations in the Asian agrifood economy: Food

security implications. Global Food Security, 3(2), pp.108-117.

Reardon, T., Chen, K., Minten, B. and Adriano, L., 2012. The quiet revolution in staple food value chains: Enter

the dragon, the elephant, and the tiger. Asian Development Bank.

Reardon, T., Fall, A.A., Kelly, V., Delgado, C., Matlon, P., Hopkins, J. and Badiane, O., 1994. Is income

diversification agriculture-led in the West African Semi-Arid Tropics? The nature, causes, effects, distribution,

and production linkages of off-farm activities. Economic policy experience in Africa: What have we learned,

pp.207-230.

Rehima, M., Belay, K., Dawit, A. and Rashid, S., 2013. Factors affecting farmers’ crops diversification:

Evidence from SNNPR, Ethiopia. International Journal of Agricultural Sciences, 3(6), pp.558-565.

Ronner, E. and Giller, K.E., 2013. Background information on agronomy, farming systems and ongoing projects

on grain legumes in Tanzania. N2Africa Milestones. Accessed, 17, pp.09-13.

Schultz, T.W., 1953. The] economic organization of agriculture(No. HD1411 S43).

Semykina, A. and Wooldridge, J.M., 2010. Estimating panel data models in the presence of endogeneity and

selection. Journal of Econometrics, 157(2), pp.375-380.

Semykina, A. and Wooldridge, J.M., 2013. Estimation of dynamic panel data models with sample

selection. Journal of Applied Econometrics, 28(1), pp.47-61.

Sen, A.K., 1966. Peasants and Dualism with or without Surplus Labor. Journal of political Economy, 74(5),

pp.425-450.

Shanin, T., 1986. Chayanov message: Illuminations, miscomprehensions and the contemporary “Development

theory”.

Shilpi, F., and Shahe, E. 2016. "Agricultural productivity and non-farm employment: evidence from

Bangladesh."

Shucksmith, D.M., Bryden, J., Rosenthall, P., Short, C. and Winter, D.M., 1989. Pluriactivity, farm structures

and rural change. Journal of Agricultural Economics, 40(3), pp.345-360.

Sigurdson, J. and Kim, Y.C., 1981. Relationship between Farm Mechanization, Rural In-Industrialization and

Saemaul Movement in Korea.


87

Singh, I., Squire, L. and Strauss, J., 1986. A survey of agricultural household models: Recent findings and

policy implications. The World Bank Economic Review, 1(1), pp.149-179.

Singh, I., Squire, L. and Strauss, J., 1986. Agricultural household models: Extensions, applications, and policy.

Singha, A.K., Baruah, M.J., Bordoloi, R., Dutta, P. and Saikia, U.S., 2011. Analysis on influencing factors of

technology adoption of different land based enterprises of farmers under diversified farming system. Journal of

agricultural science, 4(2), p.139.

Sitko, N. J., and Jayne, T. S. (2014). Structural transformation or elite land capture? The growth of “emergent”

farmers in Zambia. Food Policy, 48, 194-202.

Smith, A., 1937. The wealth of nations [1776].

Tanaka, O. 1951 “An Analysis of Economic Behavior of the Farm Household.” Journal of Rural Economics

22(4) (in Japanese).

Taylor, J.E. and Adelman, I., 2003. Agricultural household models: genesis, evolution, and extensions. Review

of Economics of the Household, 1(1), pp.33-58.

The OECD. (2016). OECD data. [online] Available at: https://data.oecd.org/ [Accessed 17 Jul. 2017].

The World Bank LSMS. http://microdata.worldbank.org/index.php/catalog/lsms.

Timmer, C.P., 1988. The agricultural transformation. Handbook of development economics, 1, pp.275-331.

Timmer, C.P., 2002. Agriculture and economic development. Handbook of agricultural economics, 2, pp.1487-

1546.

Timmer, C.P., 2017. The impact of supermarkets on nutrition and nutritional knowledge: a food policy

perspective. In Nutrition and Health in a Developing World (pp. 737-751). Springer International Publishing.

Tobin, J., 1958. Estimation of relationships for limited dependent variables. Econometrica: journal of the

Econometric Society, pp.24-36.

Traub, L., Yeboah, F., Meyer, F. and Jayne, T.S., 2015, September. Megatrends and the Future of African

Economies. In Presentation at the 2nd Annual ReNAPRI Stakeholder Conference “Anticipating the Future of

Agriculture in Eastern and Southern Africa: Outlook for Maize, Wheat, Rice and Sugar” Maputo, Mozambique.

Turner, E.C., 2014. Determinants of crop diversification among mozambican smallholders: Evidence from

household panel data. Michigan State University.

Ullah, R., Shivakoti, G.P. and Ali, G., 2015. Factors effecting farmers’ risk attitude and risk perceptions: the

case of Khyber Pakhtunkhwa, Pakistan. International journal of disaster risk reduction, 13, pp.151-157.

UNEP, (2017). [online] Available at:

https://www.unep.org/greeneconomy/sites/unep.org.greeneconomy/files/uganda_final_report.pdf [Accessed 1

Aug. 2017].

Usaid.gov. (2016). Press Releases | Uganda | U.S. Agency for International Development. [online] Available at:

https://www.usaid.gov/uganda/newsroom/press-releases [Accessed 1 Sep. 2017].

Van Dusen, M.E. and Taylor, J.E., 2005. Missing markets and crop diversity: evidence from

Mexico. Environment and Development Economics, 10(4), pp.513-531.

Van Dusen, M.E., 2000. In situ conservation of crop genetic resources in the Mexican milpa system (Doctoral

dissertation, University of California, Davis).

Velavan, C. and Balaji, P., 2012. Crop Diversification in Tamil Nadu-A Temporal Analysis. Agricultural

Situation in India, 78(12), pp.655-658.

Vyas, V.S., 1996. Diversification in agriculture: concept, rationale and approaches. Indian Journal of

Agricultural Economics, 51(4), p.636.


88

Worldatlas, (2016). [online] Available at: http://htt://www.worldatlas.com/aatlas/world.htm [Accessed 6 Aug.

2017].

Worldbank.org. (2017). Uganda Home. [online] Available at: http://www.worldbank.org/en/country/uganda

[Accessed 1 Sep. 2017].

Worldometers.info. (2017). Uganda Population (2017, 2018) - Worldometers. [online] Available at:

http://www.worldometers.info/world-population/uganda-population/ [Accessed 9 Jul. 2017].


Documents

TOWARDS AGRICULTURAL TRANSFORMATION: FACTORS INFLUENCING